EVALUATION OF DATA COLLECTION TECHNIQUES & METHODS

FOR ROADSIDE STATION ORIGIN-DESTINATION STUDIES

A Thesis
Submitted to the Faculty
of
Purdue University
by
Bryan P. Guy

In Partial Fulfillment of the

Requirements for the Degree
of
Master of Science in Civil Engineering

December 2005
 ii

ACKNOWLEDGMENTS

There are a few people I would like to thank who contributed to this project and
assisted me greatly throughout the course of its completion.
First, I’d like to thank my fiancée, Steph, for respecting my wishes to pursue this
degree, for her strength and support during our geographical separation, and for her
never-ending patience and encouragement in listening to my thoughts and concerns. I
am excitedly looking forward to beginning a new chapter in our lives.
I’d also like to express my thanks to Dr. Fricker for offering his knowledge and
expertise when I often had a lack of it. I appreciate the freedom he gave me in the
approach I took and pace at which I worked on this project. I greatly enjoyed working
with him and will leave here better prepared for whatever lies ahead.
Thanks to Nagasayan for allowing me to borrow his equipment and his
assistance provided throughout the project. I hope I returned everything to you that is
yours, and I wish you luck in your future endeavors.
Finally, thanks to everyone else, including Karen and DJ in the JTRP office, my
advisory committee, my family, and my fellow graduate students and friends who made
my experience here a very enjoyable one.
 iii

TABLE OF CONTENTS

Page
LIST OF TABLES........................................................................................................... VII
ABSTRACT ...................................................................................................................... X
CHAPTER 1 – INTRODUCTION ......................................................................................1
1.1 HOUSEHOLD TRAVEL SURVEYS (TRAVEL DIARIES) ............................................... 5
1.1.1 Example: Building a Travel Demand Model ................................................. 6
1.1.2 Example: Updating a Travel Demand Model................................................ 7
1.2 ROADSIDE STATION SURVEYS ............................................................................. 7
1.2.1 Example: Bypass Feasibility Study............................................................... 8
1.2.2 Example: Traffic Signal Re-timing ................................................................ 8
1.2.3 Example: Crash Analysis.............................................................................. 8
1.3 EMPLOYER AND SPECIAL GENERATOR TRAVEL SURVEYS ..................................... 8
1.3.1 Example: Central Business District Congestion ........................................... 9
1.3.2 Example: Airport Access .............................................................................. 9
1.4 COMMERCIAL VEHICLE TRAVEL SURVEYS ............................................................ 9
1.4.1 Example: Commercial Vehicle Travel Demand Modeling .......................... 10
1.5 ON-BOARD TRANSIT SURVEYS ........................................................................... 10
1.5.1 Example: Redesigning Transit Routes ....................................................... 10
1.6 HOTEL & VISITOR SURVEYS............................................................................... 10
1.6.1 Example: Tourism District........................................................................... 11
1.7 PARKING SURVEYS ........................................................................................... 11
1.7.1 Example: Parking Shortage........................................................................ 11
1.8 RESEARCH MOTIVATION .................................................................................... 11
CHAPTER 2 – DESCRIPTION OF DATA COLLECTION TECHNIQUES & METHODS 13
2.1 LICENSE PLATE MATCHING TECHNIQUE ............................................................. 13
2.1.1 The Clipboard Method ................................................................................ 18
2.1.2 The Audio Method ...................................................................................... 19
2.1.3 The Laptop Method .................................................................................... 20
2.1.4 The Video Method ...................................................................................... 21
2.1.5 The Photography Method ........................................................................... 23
2.1.6 Summary of Methods for License Plate Matching Technique .................... 23
2.2 OTHER (NON-LICENSE PLATE) MATCHING TECHNIQUES ..................................... 24
2.2.1 Lights-on Survey......................................................................................... 24
2.2.2 Automatic Vehicle Identification at Toll Stations......................................... 26
2.2.3 Video Imaging ............................................................................................ 27
2.2.4 Loop Detectors ........................................................................................... 27
2.2.5 Traffic Signal Preemption Devices ............................................................. 29
 iv

Page

2.3 LICENSE PLATE FOLLOW-UP SURVEY TECHNIQUE.............................................. 29

2.4 VEHICLE INTERCEPT SURVEY TECHNIQUE .......................................................... 32
2.4.1 Roadside Interviews ................................................................................... 34
2.4.2 Postcard Questionnaires ............................................................................ 35
2.4.3 Tag-on-Vehicle Survey ............................................................................... 36
2.4.4 Summary of Methods for the Vehicle Intercept Survey Technique ............ 37
2.5 VEHICLE TRACING TECHNIQUE .......................................................................... 37
2.5.1 Global Positioning System (GPS) Tracing.................................................. 38
2.5.2 Wireless Phone Tracing ............................................................................. 39
2.5.3 Summary of Methods for the Vehicle Tracing Technique........................... 41
2.6 MATHEMATICAL OD ESTIMATION TECHNIQUES ................................................... 42
2.7 SUMMARY OF ALL OD TECHNIQUES ................................................................... 42
CHAPTER 3 – STATE DOT SURVEYS ON OD TECHNIQUES AND METHODS.........44
3.1 SURVEY ON ROADSIDE INTERVIEWS ................................................................... 44
3.2 SURVEY ON ALL OD TECHNIQUES AND METHODS ............................................... 45
3.2.1 Results ....................................................................................................... 45
3.2.2 Conclusions ................................................................................................ 47
CHAPTER 4 – LEGAL ISSUES OF VEHICLE INTERCEPT AND LICENSE PLATE
FOLLOW-UP SURVEYS IN INDIANA.......................................................................48
CHAPTER 5 – EVALUATION OF EQUIPMENT FOR LICENSE PLATE DATA
COLLECTION ...........................................................................................................49
5.1 CLIPBOARD TECHNOLOGY & EQUIPMENT ........................................................... 49
5.2 AUDIO TECHNOLOGY ......................................................................................... 50
5.2.1 Analog Cassette Audio Recorders ............................................................. 50
5.2.2 Digital Audio Recorders.............................................................................. 50
5.3 AUDIO EQUIPMENT ............................................................................................ 51
5.3.1 Sony Micro-cassette Audio Recorder ......................................................... 51
5.3.2 Olympus Digital Audio Recorder ................................................................ 52
5.4 EVALUATION OF AUDIO EQUIPMENT ................................................................... 53
5.4.1 Audio Quality .............................................................................................. 53
5.4.2 Voice-Operated Recording ......................................................................... 53
5.4.3 Storage and Power..................................................................................... 54
5.5 LAPTOP TECHNOLOGY & EQUIPMENT ................................................................. 55
5.6 DIGITAL VIDEO TECHNOLOGY ............................................................................ 55
5.6.1 Resolution .................................................................................................. 56
5.6.2 Recording Media ........................................................................................ 56
5.6.3 Playback & Editing ..................................................................................... 56
5.6.4 Charge-Couple Device (CCD) .................................................................... 57
5.6.5 Magnification (Zoom).................................................................................. 57
5.6.6 Focus.......................................................................................................... 57
5.6.7 Shutter Speed ............................................................................................ 58
5.6.8 Exposure/Aperture ..................................................................................... 58
5.6.9 Other Features ........................................................................................... 58
5.7 DIGITAL VIDEO EQUIPMENT ............................................................................... 59
 v

Page

5.8 EVALUATION OF DIGITAL VIDEO EQUIPMENT....................................................... 60

5.8.1 Typical Roadside & Overhead Setup ......................................................... 60
5.8.2 Lighting Conditions ..................................................................................... 62
5.8.3 Focus.......................................................................................................... 64
5.8.4 Obstruction & Minimum Shooting Angles ................................................... 65
5.8.5 Magnification and Angle-of-View Variations by Camcorder Model............. 68
5.8.6 Magnification .............................................................................................. 72
5.8.7 Optimal Shutter Speed ............................................................................... 73
5.8.8 Left Lane Obstruction ................................................................................. 75
5.9 PHOTOGRAPHY EQUIPMENT .............................................................................. 76
CHAPTER 6 – DETERMINING LICENSE PLATE DATA COLLECTION METHOD .......78
6.1 LICENSE PLATE DATA COLLECTION RULES......................................................... 78
6.1.1 License Plate Matching Technique............................................................. 78
6.1.2 License Plate Follow-Up Survey Technique............................................... 79
6.2 LICENSE PLATE VARIATIONS .............................................................................. 80
6.2.1 Indiana Passenger Vehicle License Plates ................................................ 80
6.2.2 Indiana Commercial & Government Vehicle License Plates ...................... 82
6.2.3 Out-of-State License Plates ....................................................................... 83
6.3 LICENSE PLATE IDENTIFICATION IN THE FIELD .................................................... 84
6.3.1 Time-Length of License Plate Legibility...................................................... 87
6.3.2 License Plate Identification Time................................................................ 90
6.3.3 Probability of String Identification ............................................................... 90
6.3.4 Maximum Vehicle Speed for Manual License Plate Data Collection.......... 92
6.4 LICENSE PLATE RECORDING.............................................................................. 94
6.4.1 License Plate Recording Time.................................................................... 94
6.4.2 Vehicle Interarrival Times (Headways)....................................................... 95
6.4.3 Maximum Flow Rate for Manual Methods .................................................. 95
6.5 VIDEO METHOD ................................................................................................. 97
6.6 ERRORS IN LICENSE PLATE DATA ...................................................................... 97
6.6.1 Field and Transcription Errors .................................................................... 97
6.6.2 Transcription Time.................................................................................... 100
CHAPTER 7 – ACCURACY OF OD STUDY TECHNIQUES........................................102
7.1 LICENSE PLATE MATCHING TECHNIQUE ........................................................... 102
7.1.1 Simulation................................................................................................. 103
7.1.3 Results ..................................................................................................... 105
7.2 VEHICLE INTERCEPT AND LICENSE PLATE FOLLOW-UP SURVEY TECHNIQUES ... 112
7.3 VEHICLE TRACING TECHNIQUE ........................................................................ 114
7.3.1 Simulation................................................................................................. 115
7.3.2 Results ..................................................................................................... 117
CHAPTER 8 – GUIDELINES FOR SELECTING AN OD TECHNIQUE AND METHOD
120
8.1 SUMMARY OF FINDINGS FOR LICENSE PLATE DATA COLLECTION ...................... 120
8.2 SELECTING AN OD TECHNIQUE........................................................................ 122
8.3 SELECTING A METHOD .................................................................................... 123
 vi

Page
`
8.3.1 Accuracy................................................................................................... 123
8.3.2 Cost Items ................................................................................................ 124
8.4 CONCLUSIONS AND FUTURE RESEARCH........................................................... 125
LIST OF REFERENCES ...............................................................................................126
APPENDIX ...........................................................ERROR! BOOKMARK NOT DEFINED.
 vii

LIST OF TABLES

Table Page
Table 1: Summary of Characteristics for License Plate Data Collection Methods ..........24
Table 2: Summary of Characteristics for Vehicle Intercept Survey Methods ..................37
Table 3: Summary of Characteristics for Vehicle Tracing Methods ................................42
Table 4: Summary & Comparison of Roadside Station OD Study Techniques ..............43
Table 5: Summary of Camcorder Features .....................................................................59
Table 6: Predicted vs. Actual Angles-of-View .................................................................71
Table 7: Perpendicular Vehicle Speed (mph) .................................................................74
Table 8: Theoretical Time-Length of License Plate Legibility (2.75” Characters)............88
Table 9: Theoretical Time-Length of License Plate Legibility (1.25” Characters)............89
Table 10: Observed Time-Length of License Plate Legibility (1.25” Characters)............89
Table 11: Identification Time (s) by Vehicle Speed .........................................................90
Table 12: Vehicle Speed (mph) above which Video Method is Recommended..............93
Table 13: Recording Time (s) by Method........................................................................95
Table 14: Maximum Flow Rate (vphpl) using Manual Recording Methods .....................96
Table 15: Field & Transcription Errors by Method...........................................................98
Table 16: Transcription Time (sec) by Method..............................................................101
Table 17: Standard Deviation of Percent Error in Trip Estimations...............................112
Table 18: Actual OD Matrix (OD1) ................................................................................115
Table 19: Sample Probe Vehicle OD Matrix* ................................................................115
Table 20: Estimated OD Matrix (Expanded Probe Vehicle Matrix) ...............................116
Table 21: 4x4 1600-trip Matrix (OD2) without Uniformly-Distributed Cells....................117
Table 22: 4x4 1600-trip Matrix (OD3) without Uniformly-Distributed Cells or Zeros .....117
Table 23: 4x4 16,000-trip Matrix (OD4) with Uniformly-Distributed Cells......................117
Table 24: 8x8 1600-trip Matrix (OD5) with Uniformly-Distributed Cells.........................118
Table 25: General Cost Items by OD Study Technique ................................................124
 viii

LIST OF FIGURES

Figure Page
Figure 1: Traffic Analysis Zones and Types of Trips .........................................................2
Figure 2: Sample Origin-Destination Matrix ......................................................................3
Figure 3: Bypass Analysis for Small Cities & Towns.........................................................4
Figure 4: Corridor Origin-Destination Analysis ..................................................................4
Figure 5: Types of Trips obtained from a Household Travel Survey .................................6
Figure 6: Illustration of License Plate Matching ..............................................................14
Figure 7: Types of Trips obtained from the License Plate Matching Technique .............15
Figure 8: Proportion of spurious matches for various block sizes & permutations..........17
Figure 9: Lights-On Method ............................................................................................25
Figure 10: Typical Layout of Toll Station .........................................................................26
Figure 11: Vehicle Signatures for Various Vehicle Types ...............................................28
Figure 12: Types of Trips from License Plate Follow-Up Survey Technique ..................31
Figure 13: Typical Layout of a Vehicle Intercept Station .................................................33
Figure 14: Handset-Based Location Determination using GPS ......................................40
Figure 15: Network-Based Location Determination using TDOA ....................................40
Figure 16: Plan View of Typical Camcorder Roadside Setup .........................................60
Figure 17: Elevation View of Typical Camcorder Overhead Setup .................................61
Figure 18: Lighting Conditions Encountered on the Roadside ........................................63
Figure 19: Following-Vehicle Obstruction from the Roadside Perspective .....................66
Figure 20: Following-Vehicle Obstruction from the Overhead Perspective.....................66
Figure 21: Obstruction Angle from Roadside Perspective ..............................................67
Figure 22: Obstruction Angle from Overhead Perspective..............................................67
Figure 23: Setup on Curves in the Road.........................................................................68
Figure 24: Dimensions of Typical Camcorder Lens ........................................................69
Figure 25: Illustration of Angles-of-View .........................................................................70
Figure 26: Relationship between Focal Length & Angle-of-View by CCD Size...............71
 ix

Page

Figure 27: Required Magnification of License Plates ......................................................73

Figure 28: Vehicle Speed Perpendicular to the Camcorder Line-of-Sight.......................74
Figure 29: Optimal Shutter Speed...................................................................................75
Figure 30: Obstruction of License Plates in Left Lane ....................................................76
Figure 31: Left Lane License Plate Obstruction Factor ...................................................76
Figure 32: Indiana Passenger Vehicle Standard License Plates ....................................81
Figure 33: Indiana Passenger Vehicle Special Recognition License Plates ...................81
Figure 34: Indiana Commercial & Government Vehicle License Plates..........................82
Figure 35: Out-of-State Standard Passenger Vehicle License Plates.............................83
Figure 36: Out-of-State Commercial Vehicle License Plates ..........................................84
Figure 37: License Plate Legibility Distance ...................................................................85
Figure 38: Definition of 20/20 Visual Acuity ....................................................................88
Figure 39: Probability of 4-Character Identification .........................................................91
Figure 40: Probability of Full String Identification ............................................................92
Figure 41: Capture Rates by Vehicle Flow for each Method...........................................96
Figure 42: Error by Proportion for Traffic Volume of 100 Vehicles................................105
Figure 43: Error by Proportion for Traffic Volume of 1,000 Vehicles.............................106
Figure 44: Error by Proportion for Traffic Volume of 5,000 Vehicles.............................106
Figure 45: Error by Proportion for Traffic Volume of 10,000 Vehicles...........................107
Figure 46: Error by Traffic Volume for a True Proportion of 1%....................................108
Figure 47: Error by Traffic Volume for a True Proportion of 5%....................................108
Figure 48: Error by Traffic Volume for a True Proportion of 25%..................................109
Figure 49: Error by Traffic Volume for a True Proportion of 50%..................................109
Figure 50: Error by Capture Rate Product for Traffic Volume of 100 Vehicles..............110
Figure 51: Error by Capture Rate Product for Traffic Volume of 1,000 Vehicles...........110
Figure 52: Error by Capture Rate Product for Traffic Volume of 5,000 Vehicles...........111
Figure 53: Error by Capture Rate Product for Traffic Volume of 10,000 Vehicles.........111
Figure 54: PRMSE of Estimated OD Matrices based on OD1 (Table 18).....................116
Figure 55: Mean PRMSE for Various OD Matrices .......................................................118
 x

ABSTRACT

Guy, Bryan P. MSCE, Purdue University, December, 2005. Evaluation of Data

Collection Techniques and Methods for Roadside Station Origin-Destination Studies.
Major Professor: Jon D. Fricker.

Origin-Destination studies are often used in transportation planning to determine the

travel patterns (origin-destination matrix) of vehicles and goods in a particular area.
Given these travel patterns, the impacts of alternative solutions to current and future
transportation problems can be evaluated. Therefore, it is important that the travel
patterns be accurately measured. However, it is not always clear what data collection
method should be used to obtain the type of data needed, while maximizing quality and
minimizing the time and cost. The objective of this research is to review both
conventional and experimental techniques for roadside station OD studies, and make
general recommendations for the best OD study technique and data collection method,
given the roadway characteristics and traffic conditions.
 1

CHAPTER 1 – INTRODUCTION

Origin-Destination (OD) Studies are an important tool for transportation

professionals. OD studies are conducted to understand the pattern of the movement of
persons and goods in a particular area of interest during a particular period of time
(Wang, 1997).
OD studies are typically conducted in order to collect data as a basis for travel
demand modeling. At the core of travel demand modeling is the OD matrix, which is
essentially the trip distribution step of the four-step modeling process. Once travel
demand models are created and calibrated for an area, they can be used to perform a
variety of tasks. These tasks include analysis of travel characteristics (such as travel
time, delay, and pollution), the impact of modification or closure (due to an incident or
road work) of existing routes, and the design and evaluation of the effectiveness of new
routes on the existing transportation network. They can also be used to make long
range travel forecasts to identify potential future problems in the transportation network
and evaluate alternative solutions. Precise and accurate travel demand models are
necessary for good transportation planning and programming.
The OD matrix is an n x n matrix, where n is the number of Traffic Analysis
Zones (TAZs) in a study area plus the number of entry/exit nodes (external stations) that
lie at the cordon line (or boundary) of the study area. The size and number of TAZs
depends upon the population, employment, and land use in the area.
Figure 1 illustrates an area with 12 TAZs and four entry/exit nodes. There are
five arrows drawn on the figure, which represent the types of trips that are required for a
complete OD matrix.
 2

Figure 1: Traffic Analysis Zones and Types of Trips

An internal-internal (I-I) trip is a trip that has its origin and destination inside the
study area. Usually, these trips originate in one TAZ and are destined for another.
However, a special type of I-I trip, the intra-zonal trip, is one that has its origin and
destination within the same TAZ. An internal-external (I-E) trip is a trip that originates
inside the study area, travels through an exit node on the cordon line, and has a
destination outside the study area. On the other hand, an external-internal (E-I) trip is
one that originates outside the study area, travels through an entry node on the cordon
line, and is destined for some TAZ inside the study area. Finally, an external-external
(E-E) trip is a trip that has both its origin and destination outside the study area, but
passes through the study area via two entry and exit nodes.
In order to obtain a complete OD matrix, the number of trips from each TAZ (and
entry/exit node) to all other TAZs (and entry/exit nodes) must be determined. Therefore,
the OD matrix for Figure 1 is a 16x16 matrix (12 TAZs plus 4 entry/exit nodes). In
reality, however, OD matrices are much larger, because metropolitan areas may contain
hundreds of TAZs.
 3

Figure 2 illustrates the OD matrix for the area shown in Figure 1. To illustrate the
location of the types of trips described above, the cells are shaded and labeled. In
reality, each cell would contain a number that represents the number of trips from the
origin zone to the destination zone. The sum of each row represents the total number of
trips that originate in each zone, and the sum of the column represents the number of
trips that have destinations in each zone.

Figure 2: Sample Origin-Destination Matrix

While complex travel demand models are typically developed for large cities,
metropolitan planning organizations (MPOs), and state departments of transportation
(DOTs), smaller cities and towns without a complete travel demand model sometimes
also require OD studies in order to evaluate alternative solutions to transportation
problems. These travel demand models can be used to evaluate bypass alignment
alternatives, traffic signal coordination and timing scenarios, and corridor safety, among
others.
Figure 3 illustrates a bypass analysis for a small city or town. The objective in
this situation is to determine the number of E-E trips. The principles are the same as
those for a larger study area; however, less attention is given to I-I trips because the
 4

entire study area is considered to be one or a few TAZs. E-I (and I-E) trips are
determined by subtracting the traffic volumes on the link at the entry/exit node from the
number of E-E trips determined at that location. This process is conducted at all
entry/exit nodes. In the figure, the thickness of the arrow represents the relative traffic
flow from only one entry node to all other exit nodes (E-E trips) and the city itself (E-I
trips).

Figure 3: Bypass Analysis for Small Cities & Towns

Figure 4 graphically illustrates how an OD study might be applied along a

corridor for signal timing coordination or safety analysis. Again, the thickness of the
arrow represents the relative traffic flow.

Figure 4: Corridor Origin-Destination Analysis

 5

Other, more specialized OD studies can also be used to analyze interchange

design, special traffic generators such as airports and other large tourist attractions, and
parking adequacy. Typically, an OD study can be categorized into one of seven types.
The following is a description of each type along with an example where each is
commonly applied.

1.1 Household Travel Surveys (Travel Diaries)

In this survey, a sample of households within the study area (with varying levels
of income, household members, and available vehicles) are selected and recruited for
participation. Members of participating households record their household information
and travel activity in a travel diary or through a recall interview for a certain time period,
generally 24 or 48 hours. The information recorded includes the start time, travel time,
trip length, origin, destination, travel mode, trip purpose, and vehicle occupancy of each
trip. This information is aggregated with other similar households to establish average
trip rates and trip lengths for that type of household. This data is expanded to all
households in the study area. By combining this information with trip production and
attraction analysis in the first step of the travel demand modeling process, an origin-
destination matrix can be defined. This type of survey is typically conducted by MPOs
as an update to an existing travel demand model, or by the US Department of
Transportation Bureau of Transportation Statistics as part of the National Household
Travel Survey (NHTS).
Besides administering the survey to a new subset of households for each survey,
a panel of households may be used. In a panel sample, the travel survey is given to the
same households over time, thus revealing changes in travel behavior as household
characteristics (such as the ages of household members) change.
The household travel survey reveals general travel patterns from a sample of the
population for all roadways in a particular study area. It does not provide detailed
information on any particular roadway within the study area. Figure 5 illustrates the
types of trips that are obtained as a result of conducting a household travel survey.
 6

Figure 5: Types of Trips obtained from a Household Travel Survey

Map Source: Box & Oppenlander as shown in the Manual of Transportation
Engineering Studies, 1994

The arrows in Figure 5 represent the traffic flows from one TAZ to all other TAZs.
The dark-shaded arrows represent inter-zonal I-I trips, while the lighter-shaded arrows
represent I-E trips. E-E trips can not be obtained from a household travel survey.
The following examples are some common applications where household travel
surveys are generally used.

1.1.1 Example: Building a Travel Demand Model

A small metropolitan area is one of the fastest growing areas in the state, and it
is expected that it will continue to grow for the foreseeable future. Many citizens are
 7

upset that road construction is not keeping up with the increased demand on the
roadways, and are demanding better short-term and long-term planning strategies. The
new MPO in charge of the area’s comprehensive plan wants to develop a travel demand
model to represent the existing transportation network and forecast future growth so that
road construction can be adequately planned and designed in advance of the demand.
To do this, a household travel survey will be administered to a sample of all households,
and certain households will be recruited to complete a 24-hour travel diary.

1.1.2 Example: Updating a Travel Demand Model

A major update of an urban area’s travel demand model is not going well. The
model’s link loadings do not match the observed counts on the corresponding links at all.
The mismatches are especially bad on the high-volume links. After ruling out other
causes, the modeler suspects the trip table used in the trip distribution step. The table
was an update of an origin-destination matrix that was based on a household survey
conducted 25 years ago. The update was based on a borrowed equation for the
average trip lengths (by trip purpose) in unspecified cities of similar population, but the
modeler suspects that the highways that pass through his city cause the borrowed
equation to produce erroneous results.

1.2 Roadside Station Surveys

In this type of OD study, drivers are directly interviewed or vehicles are monitored
(typically via license plates) at a set of roadside stations to determine their travel
characteristics through the study area. This survey method is used to determine the
number of E-E trips on each road segment at the entry/exit node. Depending on the
technique used in this study, I-E and E-I trips may or may not be determined. This will
be discussed further in Chapter 3.
Like the household travel survey, this type of study has many uses. The
roadside station survey is often used to supplement the household travel survey in
creating and validating travel demand models by collecting data at important internal
cordons and screen-lines. It can also be used to collect data in a particular corridor for
sub-area and small area models. Unlike the data from household travel surveys, data
from roadside station surveys are typically not transferable to other applications. The
following examples describe some common applications of the roadside station survey.
 8

1.2.1 Example: Bypass Feasibility Study

A small city lies at the intersection of two state highways. In recent years, truck
traffic through the small city seems to have increased greatly. The city’s residents
suspect that most of the heavy trucks do not have origins or destination in the city; they
are just passing through. In doing so, the trucks are destroying the pavement on the
city’s two main streets and causing bad traffic tie-ups at the city’s principal intersection.
If it can be proved that most truck traffic is through traffic, a strong case for building a
bypass around the city can be made.

1.2.2 Example: Traffic Signal Re-timing

A major US highway passes through a medium-sized city, which also serves as a
major arterial in the city’s street network. The traffic signals along that highway corridor
are timed as if most of the traffic is through traffic. Some people think that much of the
traffic on that highway has at least one trip end in the city – at least during peak periods.
If that is true, the city’s travel demand model must be revised to reflect those trip
patterns and the corridor’s traffic signals should be retimed to accommodate the turns on
and off the highway.

1.2.3 Example: Crash Analysis

A particular section of urban interstate highway in a major city has an unusually
high crash rate. Many of these crashes seem to be sideswipe and rear-end collisions.
The highway section is located between two entry/exit ramps that are rather close to
each other. Some observers think that weaving is a problem. If the pattern with which
vehicles enter and exit the interstate at these two locations can be established, the
expensive redesign of those two entry/exit points can be evaluated.

1.3 Employer and Special Generator Travel Surveys

This type of survey is used to collect information on establishments with trip
attractions. These establishments may be businesses or other unique special
generators such as commercial airports, arenas and convention centers, amusement
parks, or other tourist attractions. This type of survey is often conducted because the
establishments have high attraction rates that have unique characteristics and cover a
 9

large geographic area. If the survey is being conducted for one or a few establishments,
the survey can be conducted as an intercept survey as people enter and exit a building
or site similar to an roadside station survey. Otherwise, to gather information on a large
number of establishments, the survey may be distributed to a sample of the population
at each establishment, similar to a household travel survey. The following are some
examples where employer or special generator travel surveys may be used.

1.3.1 Example: Central Business District Congestion

During the afternoon rush-hour in the CBD of a large city, traffic congestion at the
freeway on-ramps create long delays and is a safety concern of drivers and pedestrians.
In an effort to reduce congestion during this time, surveys were sent out to major
employers in the CBD to determine if any alternatives (such as alternate routes,
staggered business hours, or transit subsidies) would reduce delays and increase
safety.

1.3.2 Example: Airport Access

Annual passenger enplanements at an airport in a large city have grown steadily
the past several years. Traffic congestion at the airport’s loading and unloading area at
the terminal is creating long delays and is a safety hazard for pedestrians. The city and
airport are considering constructing a new light-rail line connecting the central business
district about three miles away to the inside of the airport terminal. It is important to
determine the percentage of airport patrons coming and going from the CBD, and just as
important, where within the CBD. Therefore, a survey will be given to patrons, and for
those with trip ends in the CBD, a stated preference survey will be administered to
evaluate use of the light rail line.

1.4 Commercial Vehicle Travel Surveys

This survey type is used to obtain OD data for trucks and other commercial
vehicles. This information can be utilized to determine trip rates, commodity flows, and
air quality modeling within an area. The following is one example where a commercial
vehicle travel survey may be used.
 10

1.4.1 Example: Commercial Vehicle Travel Demand Modeling

In recent years, a metropolitan planning organization has been updating its travel
demand model. Because most household travel surveys do a poor job of defining
commercial trips, and traffic counts cannot distinguish between a personal trip and
commercial trip for light-duty vehicles, a new category of trip is being created to be
included in the updated travel demand model which will better define those trips.
Therefore, data will need to be collected on the travel patterns of vehicles such as
package delivery vehicles, postal vehicles, couriers, equipment repair and service
technicians, craftsmen (carpenters, plumbers, etc.), government workers, and taxis, all
of which may use personal or light-duty vehicles for commercial purposes.

1.5 On-board Transit Surveys

This type of survey is used by modelers or transit agencies. Information is
typically collected on characteristics of the users and their travel patterns. The transit
origin-destination matrix can be used in travel demand models or by transit agencies to
analyze service changes and improve the performance of the transit system. The
following is an example of one application of an on-board transit survey.

1.5.1 Example: Redesigning Transit Routes

Over the course of the last several years, the ridership on a city’s bus system has
declined. At the same time, much of the new commercial development has occurred on
the city’s edge. In an effort to expand to the new development, the bus company has
extended the existing routes into the new areas, which has caused more stops and
delay for all of the riders. Officials want to know if breaking up these routes and adding
express routes will slow or stop the ridership decline. As part of this analysis, a survey
of passengers on-board the buses will be conducted.

1.6 Hotel & Visitor Surveys

This type of survey seeks to collect data on travel conducted by visitors that stay
in hotels and motels and who are out of reach of the household survey. Typically, these
groups of people have much different travel characteristics than the residents of the
same city. This type of OD study could be considered a type of special generator survey
 11

described in Section 1.3. The following is an example of one application of a hotel/visitor

travel survey.

1.6.1 Example: Tourism District

A particular tourist district of a city contains a mixture of land uses, including
entertainment, shopping, and recreational establishments, hotels, and an amusement
park. The streets in and around this district are highly congested. Many of the patrons
are out-of-town visitors who are unfamiliar with the area. The city and property owners
of the district are interested in building a transit system to link the district’s top attractions
in an effort to reduce the congestion on the streets. A survey of hotel guests will be
conducted to determine the daily travel patterns of these visitors on weekdays and
weekends.

1.7 Parking Surveys

Parking surveys are sometimes used to obtain detailed parking information for
travel demand models to evaluate parking supply, costs, and subsidies on travel
decisions. This can be done as people enter and exit parking facilities or through a
questionnaire placed on the windshield of the vehicle that is voluntarily completed and
returned. The following is an example of one application of a parking survey.

1.7.1 Example: Parking Shortage

The central business district of a city is experiencing a lack of available public
parking. The city currently owns several public garages, but is considering constructing
several more. In order to determine the locations of future garages, a parking survey will
be administered to the users of existing garages to determine the type and location of
the activity for which they are being used.

1.8 Research Motivation

As indicated earlier, OD studies can take on many forms. However, all studies
require a certain amount of effort for data collection, and the quality of data can have a
major influence on the results of the study. In addition, many studies are restricted by
time and cost. Therefore, it is imperative that the amount and quality of data collected is
adequate for each and every study. It is generally not easily known, however, what data
 12

collection method should be used for a particular type of OD study because of cost
restraints, data quality, and the reliability and accuracy of the results after post-
processing.
While quite a lot of information has been published on how to best conduct
household travel surveys and travel diaries, especially in the Travel Survey Manual and
other documents published by the Travel Model Improvement Program (TMIP), less
information has been published on the best methods for conducting roadside station
surveys in regard to the data collection involved with each of those methods. The
purpose of this study is to evaluate both the conventional and experimental techniques
for conducting roadside station origin-destination studies through literature review,
surveys, field testing, and computer simulation, and subsequently develop guidelines for
conducting them. These guidelines, however, apply only to roadside station OD studies,
which are just one of the seven types of OD studies described in Chapter 1.
The Indiana Department of Transportation (INDOT), in conjunction with the
Purdue University Department of Civil Engineering, seeks to develop general guidelines
that will serve as a starting point for planning and conducting future INDOT roadside
station OD studies. These guidelines will be designed for transportation professionals,
DOTs, consultants, and others who desire to conduct a Roadside Station Origin-
Destination Study.
 13

CHAPTER 2 – DESCRIPTION OF DATA COLLECTION TECHNIQUES & METHODS

Data for origin-destination studies can be collected in many ways, particularly for
the roadside station survey. However, it is not always easy to determine which method
is the best for obtaining accurate data while minimizing cost with the resources available.
Each method has a unique set of characteristics with respect to planning, data
collection, and data analysis. The techniques and methods described in greater detail in
this chapter refer only to Roadside Station Origin-Destination Studies. To clarify, the
word “technique” refers to the procedure in which data is being collected, and “method”
refers to the means by which the procedure is carried out.

2.1 License Plate Matching Technique

In this technique, partial license plate strings are recorded by an observer at a
roadside station. In addition, the time of day is recorded from a timepiece that has been
synchronized with all other stations. Traffic volume counts are usually conducted
simultaneously (with manual traffic counters) in order to obtain a “capture rate” for
expansion of the data to the population. Vehicle classification may also be observed
and recorded for each license plate. The record of all license plates recorded from one
roadside station is compared to records from other stations within the study area. A
match is noted if a license plate is seen at two stations within a certain time parameter,
that is, upper and lower bounds of reasonable travel time between the two stations.
Figure 5 illustrates this process.
 14

Figure 6: Illustration of License Plate Matching

Source: Adapted from Slavik (1986)

Once each license plate has been evaluated for a match with all other stations,
the data is then expanded to determine the number of all vehicles at a particular station
that pass by any of the other stations, which are considered E-E or through trips. Figure
7 illustrates the types of trips that are obtained as a result of a cordon station license
plate match. The lighter-shaded arrows represent E-E trips from one entry node to all
other exit nodes on the cordon line. The single dark-shaded arrow represents all E-I
trips from the entry node to all TAZs. While the license plate matching technique can
determine the number of E-I trips, it is unknown how these E-I trips are distributed to
each TAZ. This technique is most appropriate when conducting OD studies on small
 15

cities and towns where internal trips (I-E or E-I) do not have to be assigned to a
particular TAZ (origin or destination) inside the cordon line.

Figure 7: Types of Trips obtained from the License Plate Matching Technique

The license plate matching technique has several advantages. The only data
that is recorded is the license plate, time, and if necessary, vehicle classification.
Because there are no driver surveys and vehicles are only monitored from the roadside,
this technique is unobtrusive to the drivers and safer for the observers. The data
reduction process is simpler than that of a questionnaire, but may or may not be as time-
consuming. In addition, the amount of data that can be collected for a particular road is
only limited to the means by which it is being collected. In other words, a fairly large
 16

ratio of the vehicles on the roadway can be sampled relative to that of one observer with
other techniques. According to Turner (1996), this data can provide an estimate of travel
times (if there are enough matches and precise time stamps) over the course of the
study period. Also, the technique can be used on most types of roads (Turner et al.,
1998).
However, some disadvantages exist. The weather can pose a problem
depending upon the method chosen to record license plates, and fatigue can be a
problem for observers if the data collection period is too long or too intense. Also, errors
can be introduced in several processes during the data collection and reduction periods
of identifying, recording, and transcribing each license plate. In addition, large amounts
of data can be lost due to equipment failure (such as loss of power in audio recorders,
laptops, or camcorders in the field during the study), which would require redoing parts
of the study, which is both costly and time-consuming. Locations for the roadside
stations are important, given that vehicles traveling in platoons (particularly near signals)
or low speeds may block the view of other vehicles’ license plates, especially if
camcorders or cameras are used to record them. Similarly, vehicles traveling freely at
high speeds may pass too quickly to be recorded. Furthermore, some vehicles may
have license plates that are missing, damaged, dirty, covered, or blocked. For large
surveys, adequately training and mobilizing the observers can be a challenge (Martin,
1993).
There has also been some concern about spurious matches, which are produced
when two partial license plate entries that actually belong to two different vehicles are
matched (Slavik, 1986). However, Slavik determined that the effect of spurious matches
could be virtually eliminated by updating the time at least every 5 minutes and/or
increasing the number of characters (or permutations) recorded for each entry. The
number of permutations N depends on how many characters are recorded. For three
numbers, there are 1,000 permutations (10³). 10,000 permutations are achieved by
recording four numbers (104) or three letters (22³), assuming that 22 of the 26 letters of
the alphabet are used on license plates. Figure 8 illustrates this.
 17

Figure 8: Proportion of spurious matches for various block sizes & permutations
Source: Slavik (1986)

In Figure 8, the size of the block refers to the precision with which the time
stamps are recorded for each license plate entry. Small blocks refer to precision of less
than 5 minutes; medium blocks are between 5 and 15 minutes; and large blocks are
greater than 15 minutes.
The effect of spurious matches is only an issue when large blocks are used for
any number of permutations, or for medium blocks with 1,000 permutations. In most
situations, it is not difficult to obtain small blocks. It is important, however, that
timepieces be synchronized between stations.
 18

2.1.1 The Clipboard Method

This method requires observers at each station to manually record with a paper
and pencil the partial license plate number of as many passing vehicles as possible on a
form. During intervals between vehicles, the time stamp should also be recorded (ideally
every minute but not greater than five minutes). According to Martin (1993), observers
can record approximately 170 full license plates per hour or 800 partial license plates per
hour (three or four digits).
Vehicle classification may also be recorded, but simultaneous traffic counts
should be conducted in order to obtain an accurate capture rate (which is used later for
data expansion). These records must be manually transcribed into software for
matching after the data collection period is over.
The clipboard method is the most conventional method of collecting the data. It
does not require the purchasing and supplying of any expensive equipment to the
observers. In addition, little time has to be spent on training staff.
There are disadvantages in using a clipboard, however. First of all, the license
plate entries recorded on the clipboard may be illegible to varying degrees depending
upon the quality of penmanship of each individual observer. Even one incorrectly
recorded letter or digit in the license plate string will eliminate a potential match.
Emphasis must be placed on the legibility of the penmanship of the observer. The
accuracy of each recorded license plate is more important than the quantity of license
plates recorded. If possible, the transcription of license plates should be done by the
observer. In addition, the time stamp that is recorded along with each license plate
record will not be as accurate as other methods (as will be discussed later), so the travel
times calculated from the field data may not be reliable in all situations. Weather
conditions, such as rain, can make this method difficult. Finally, according to Turner et
al. (1998) it takes a long time (approximately 10 hours per hour of data collected) to
transcribe and match the license plates.
One variation of the clipboard is the tablet PC. In recent years, tablet PCs have
become popular and less expensive. A tablet PC is virtually an electronic clipboard,
which stores the information in an electronic file rather than on paper. However, the
data recorded in the field will still have to be manually transcribed to be matched. While
handwritten character recognition technology exists to a degree in some tablet PCs and
 19

PDAs, it is likely unsuitable for recognizing quickly-written license plate strings due to the
fact that recognition accuracy is extremely important. Therefore, this was not evaluated
as part of this study.

2.1.2 The Audio Method

In this method, observers speak the partial license plate string of as many
passing vehicles as possible into some audio recording device. Like the clipboard
method, the time should be updated during intervals in which vehicles are not present.
The observer may also collect information about vehicle classification if desired. Traffic
counts must also be conducted simultaneously, or the observers can note passing
vehicles that were unidentified (by saying “no ID” into the recorder) in order to obtain a
capture rate for expansion of OD pairs to the entire population of vehicles on the road.
According to Martin (1993), between 1000 and 1200 license plates per hour can be
recorded by one observer. These voice records must then be played back after the data
collection period and manually transcribed into software for matching with other roadside
stations.
The audio method also has several advantages and disadvantages. The very
nature of audio recording allows the observer to speak, rather than write or type, the
license plate numbers into a recorder. The observer does not have to repeatedly look
up and down between the license plate and clipboard/computer screen for each entry.
This will likely allow the observer to record license plates at a faster rate than the other
methods. Audio recorders are also less susceptible in poor weather conditions than
clipboard, laptop, and video methods. Audio recording equipment is fairly inexpensive at
this time.
Depending upon the tone, pitch, and speaking rate of the observer’s voice, some
of the license plate records may be inaudible or indecipherable during the transcription
process. In addition, the noise from the traffic in the background may drown out the
observer’s voice and make the transcription process difficult as well. Mistakes made on
the audio tapes cannot be easily erased from the record in the field and may cause
confusion during the transcription process. Furthermore, like the clipboard method, time
stamps cannot be recorded as accurately as other methods. Transcription generally
takes two to three hours per hour of data collected (Turner et al., 1998). It can be
difficult to transcribe letters, because B, C, D, E G, P, T, and V, M and N, and F and S
 20

sound similar. The phonetic alphabet (alpha, bravo…) works better but requires lots of
time and resources to train staff (Martin, 1993). Higher-pitched (women’s) voices are
typically easier to understand, and it is best for the observers to transcribe their own
audio recordings if possible.
Like handwritten character recognition discussed in the previous section, speech
recognition software exists that will convert spoken words into digital characters, which
has the potential to eliminate the time-consuming process of transcribing the license
plate strings from the audio recordings to the software for matching with other stations
(Washburn et al., 1997).
The main advantage in using speech recognition software is the time saved
transcribing the license plate records manually into software. As stated above, manual
audio transcription typically takes two to three hours for every hour of tape (Turner et al.,
1998).
Speech recognition software, however, has generally been created for
recognizing spoken words, not for recognizing individual letters and numbers of a license
plate string on a noisy roadside station. Unfortunately, this may cause the automatic
transcription process to have very high error. Evaluation of this technology was not
conducted as part of this project.

2.1.3 The Laptop Method

In this method, the person at each station is equipped with a laptop computer into
which he or she types partial license plate string into the computer. The strings are
stored along with an exact time stamp (to the nearest second) assigned by the computer
after the entry is completely entered. Like the clipboard and audio methods, traffic
volume counts must be conducted to determine the number of missed vehicles. Unlike
the other methods (clipboard, audio, video, or photo), the laptop method does not
require a subsequent transcription step, because the license plate strings and time
stamps are entered directly into the computer in the field. 900 4-character strings can be
recorded per hour with this method (Turner et al., 1998).
As with the clipboard method, there are advantages and disadvantages in using
a laptop for recording license plate strings. Laptops are advantageous in data collection,
because the license plate records will all be legible, although they still may not be keyed
in correctly in the field. A time stamp can also be accurately recorded to the nearest
 21

second for each license plate, allowing precise travel times to be calculated for matched
vehicles. Furthermore, transcription is not necessary, resulting in no transcription errors
and time saved in post processing the data.
However, as with the clipboard method, license plates will be recorded incorrectly
at varying degrees depending on the typing skills of the observer. In addition, it may not
be cost effective to provide a laptop for each data collector, especially if there are many
stations in the study. In addition, laptop computers cannot be utilized in the rain. Still
another problem is providing an adequate amount of power to each laptop for the entire
study period, which could be problematic for long study periods.

2.1.4 The Video Method

This method requires observers to record the license plates on all passing
vehicles with a camcorder. The video tapes are then reviewed in the office where partial
license plate numbers, time stamps, and vehicle classification (if desired) are
transcribed. Missed vehicles also need to be counted. Vehicles can be missed (seen
but have unidentifiable license plates) due to a number of reasons, such as the
camcorder’s improper field-of-view, poor video quality (because of lighting or focus),
blockage by another vehicle, or simply the absence of the license plate on the vehicle.
Video data collection also has many advantages compared to other methods of
license plate string data collection. With video, camcorders can be set up on the side of
the road to record the license plates of passing vehicles, eliminating the need to
manually record the license plates on the roadside, especially during long study periods,
high speed, or high traffic flow conditions. The license plates can be transcribed
manually in the lab at a rate the recorder can control in order to maximize the number of
license plates recorded. Transcription typically takes 10 hours for every hour of analog
recording (Turner et al., 1998). There is also a permanent record with video, which can
be stored and referred to later if necessary. The time stamp on video can also be
accurate to the nearest second. Typically, video is the best method to use on high
speed, high flow facilities (Shuldiner, 1996).
There are some disadvantages in using video for data collection. Depending on
the study size, a large number of video camcorders may be necessary. This equipment
can be expensive and sensitive to weather conditions. It is also not well established how
to obtain the most legible license plates due to weather conditions such as sun and
 22

glare, traffic conditions such as flow rate, speed, and headways, and camera settings
such as zoom, shutter speed, exposure, and these settings may have to be changed
throughout the study period. Also, video camcorders do not necessarily eliminate
manpower, as a person may be required to be at the camcorder site to keep the camera
running (battery power and tape replacement) and to prevent theft. In addition, while a
lot of license plates can be recorded, this method still requires someone or something to
transcribe the license plate strings from the video to a computer. Like the audio method,
this process is a long, monotonous, and sometimes frustrating process.
Like the clipboard and audio methods described earlier, technology exists to
convert the license plate strings from the video automatically to digital characters. There
are a few steps in this process. First, the video has to be filtered so that one frame
containing a license plate is from each passing vehicle is saved (the others can be
removed). Second, the actual license plate has to be found within the frame. Finally,
the digits on the license plate have to be read from the video frame (Gupta et al., 2002).
The obvious and biggest advantage in using character recognition for
transcription is the time saving over manual transcription. Even if only the first step of
the process is completed (frame filtering), the time to manually transcribe the license
plate strings is greatly reduced.
The biggest disadvantage in using automatic transcription is that there may still
be some transcription error (due to the capabilities of the license plate reader and poor
quality of video). Automatic transcription typically yields fewer license plates than
manual transcription, although it can be combined with manual transcription (Shuldiner,
1996), in which case, a human tries to identify license plate characters that the machine
cannot. This technology is also relatively new, so most software is proprietary and not
readily found on the market. For these reasons, character recognition of video images
was not evaluated as part of this project. If used, however, there may be additional
constraints during the video recording process on the roadside. License plate
transcription systems are not standardized, are sensitive to ambient conditions, and can
be costly for small studies (Turner et al., 1998).
 23

2.1.5 The Photography Method

The final license plate matching technique to be discussed is still photography.
In this method, still photos are triggered by vehicles and taken of license plates as they
pass by the observer. Precise time and date stamps can also be applied to the photos.
In one Japanese application, fixed overhead cameras were able to obtain 90%
capture with 5% error of recognizable characters on freeways. The cameras can also
record color and size of the license plate (Asakura et al., 2000).
The major advantage of still photography is the reduced transcription time and
elimination of recording error. Like the video method, most automatic license plate
readers search for a video frame that contains a license plate (like a still photo) from
which the license plate number can be obtained. Essentially, the photo method
eliminates the frame-filtering process of the video method.
Few studies have been documented that use this method, and it is unknown if
this method has ever been used for an origin-destination application. However, lessons
can be learned from other applications such as systems that monitor red light running at
signalized intersections. In that application, several photos from several different
camera locations are used to obtain the license plate of red-light running vehicles so
citations can be mailed to the vehicle owner. In this situation, however, the setup is
usually permanent.
For OD purposes, the cameras would have to be able to take pictures at a high
rate (approximately every 2 seconds on average). The cameras would also need a
triggering device that would indicate when the license plate is going to be in the field-of-
view. In addition, the setup location needs to be flexible for moving from one study
location to another.
While this method was not directly evaluated as part of this project, many of the
same issues discussed for video data collection apply to this method as well. These
issues are discussed in Chapter 7.

2.1.6 Summary of Methods for License Plate Matching Technique

Table 1 below lists some of the characteristics of data collection methods for the
license plate matching technique. Each column represents one of the five data
collection methods described in this chapter. The table entries qualitatively assess
these characteristics, so that each method can be compared with the others.
 24

Table 1: Summary of Characteristics for License Plate Data Collection Methods

Clip Aud Lap Vid Photo

Equipment costs low med high high high
Observer training low low low high high
Setup difficulty low med med high high
Subject to weather yes yes yes yes yes
Power required none low high high high
Require separate traffic counts yes yes yes no yes
Precise time stamps (travel times) no no yes yes yes
Maximum traffic speeds low low low high high
Maximum traffic flows low med low high high
Recording accuracy med med med high high
Manual transcription time med med none high med
Automatic transcription accuracy med med none med med
Permanent Record of LPs no no no yes yes

2.2 Other (Non-License Plate) Matching Techniques

The techniques described below are similar to the license plate matching
technique described above. However, instead of matching the license plates of passing
vehicles, each technique tries to match some other characteristic of the vehicle. This
technique requires the same number of roadside stations as the license plate matching
technique. While each of the methods are described below along with advantages and
disadvantages, these methods were not evaluated further as part of this project.

2.2.1 Lights-on Survey

In this method, vehicle headlights (instead of license plates) are matched
between stations. To do this, the driver is simply asked to turn on his or her headlights
for the remainder of that trip, usually via roadside work zone signs or a dynamic
message sign. Then, observers at other roadside stations record the percentage of
passing vehicles with its headlights on. Figure 9 illustrates this method.
 25

Figure 9: Lights-On Method

A lights-on survey is a simple way to conduct an OD study. However, only

information about origin and destination stations can be obtained. Other information,
such as travel time and trip purpose, cannot. In addition, observers can generally
maintain a safe distance from the roadway.
Today, with the increasing use of daytime headlights on many vehicle models, a
lights-on survey could have a high rate of error. It is also unknown how many drivers
would comply with the initial request to turn on the vehicle’s headlights and subsequently
turn them off at a later station. Separate studies to determine these values would be
required before or after the study. In addition, only one origin can be studied at a time.
Therefore, for a study area that contains multiple roads, each road will have to be
studied on separate days. Finally, this technique is limited to daylight hours only.
 26

2.2.2 Automatic Vehicle Identification at Toll Stations

In many large cities across the United States, toll roads are utilized as part of the
transportation system. Many of the toll stations located on these roads are equipped
with a system that is designed to allow frequent road users to pass quickly through toll
stations without stopping in the manual toll lanes. These users typically purchase a toll
tag that is mounted on the dashboard or window of the vehicle. This toll tag has a
unique identifier that is attached to an account that contains prepaid funds by the owner
of the vehicle. When the vehicle passes through the automatic toll lanes at a toll station,
a scanner or reader obtains the identification of the vehicle via radio frequency and
deducts the toll from the account. Figure 10 illustrates how the system is set up.

Figure 10: Typical Layout of Toll Station

Source: Turner et al. Travel Time Data Collection Handbook (1998).

This type of setup could be used for OD applications. In this technique, the
observer is replaced by the toll tag scanner. Toll tag identifiers, rather than license
plates, are recorded and matched between stations. Accurate time stamps could also
be obtained for travel times. Toll systems (which often use some sort of radio frequency
technology) is very efficient at identifying vehicles. Typically, the recognition rate is
close to 100%.
 27

On the other hand, some people may consider the use of toll tag information to
be an invasion of privacy. Furthermore, only a fraction of road users have toll tags, and
those users may not be a representative sample of the population of users on the road.
For example, toll tag owners typically have a higher income and are more frequent users
of the toll system than the average population of vehicles and subsequently will have
different travel characteristics. Finally, toll tag reading technology is limited to the
roadways on which they are set up, and they cannot simply be moved to conduct a study
elsewhere.

2.2.3 Video Imaging

This type of system has been tested, although it has been used more extensively
for travel time estimation than in OD applications. Video imaging systems work by
capturing images of vehicles at different locations and matching them based on vehicle
parts such as the hood, roof, trunk, or wheels, as well as overall shape and color.
However, most of the testing was done on freeway corridors where the stations were
relatively close (Turner, 1996). The expansion of video imaging to an OD application is
likely unfeasible at this time due to the large numbers of similar vehicles that would be
recorded in a given interval.
The disadvantages seem to outweigh the advantages at this time. For example,
it would not be easy to distinguish one black Ford Taurus SE from the next on any given
roadway without recording some other unique identifier (such as a license plate
number), which is just as easily done using the video license plate matching method
discussed above.

2.2.4 Loop Detectors

Inductive loop technology is another method that has been studied in the
literature and suggested for use in an OD application (Oh et al., 2003). This is done by
utilizing a loop detector card to measure inductance, or a “vehicle signature,” that is
created by the electrical inductance of the loop as a vehicle passes over it. Vehicles of
different sizes and number of axles produce unique vehicle signatures that can be
matched among multiple loops. Figure 11 illustrates the vehicle signatures for various
vehicle types.
 28

Figure 11: Vehicle Signatures for Various Vehicle Types

Source: Turner et al. Travel Time Data Collection Handbook (1998).

This method may have certain advantages in a signalized network with closely-
spaced intersections. In those situations, it is likely that loops already exist on the
intersection approaches.
This method, however, faces many of the same problems as video imaging does.
For example, the electronic signature left by one Ford Taurus should almost be identical
to the next Ford Taurus, which limits the area over which such a method can be applied
(for example, beyond an intersection). In addition, inductive loops are expensive to
install and immobile once they are installed. Furthermore, most loops probably do not
have detector cards capable of measuring (most loops are either activated or
inactivated) and storing vehicle signature data.
 29

2.2.5 Traffic Signal Preemption Devices

It has also been suggested in the literature to utilize traffic signal preemption
devices (often found in emergency vehicles) in place of license plate observers. This
method is very similar to electronic toll tags because, like toll tags, emergency
preemptors have a unique identifier associated with them. When a signal picks up an
approaching emergency preemptor, it first has to identify it to make sure that it is a
legitimate vehicle to adjust the traffic signal.
With this technology, it would be ideal for the receiver to obtain and record the
identifier of preemption devices without actually adjusting the traffic signal to unapproved
(non-emergency) vehicles. This could then be utilized much like the technology of the
toll system.
Signals in many cities already have emergency preemption devices on their main
roadways. It is unknown, however, but unlikely that any are capable of actually
identifying and storing this type of information.
The major disadvantage is that, in order to conduct an origin destination study on
a large scale, preemption devices would have to be distributed to the drivers in the
network. This in itself makes this method highly impractical. Most cities would probably
be unwilling to use these emergency systems for this type of traffic data collection.

2.3 License Plate Follow-Up Survey Technique

This technique uses one of methods described above in order to record license
plates at a particular roadside station. A list of license plates is then supplied to the
motor vehicles department (DMV) to obtain contact information for the vehicle owner. A
survey is sent to the vehicle owner, who is then asked to respond to a survey of
questions regarding the specific trip on which their license plate was recorded.
In order to obtain contact information of vehicle owners from the DMV, the full
license plate must be recorded. Depending upon the recording method and the
requirements of each DMV, the license plates may or may not have to be transcribed
into a specific format.
Once the contact information is obtained from the motor vehicles department, a
survey of the vehicle owners can be conducted. It is critical that the date, time stamp,
location, direction of travel, and other relevant information (such as how their vehicle
 30

was recorded and contact information obtained) be included in the information provided
to the vehicle owner. This survey is usually conducted via a telephone interview or
postcard mail-out with response via mail-in, telephone, and/or internet.
License plate follow-up surveys have resulted in both successful and
unsuccessful OD studies. They are beneficial in that they are unobtrusive like the
license plate matching technique, but detailed information (trip purpose, true origin and
destination, etc.) can still be obtained from the actual driver of the vehicle using that
specific road. Figure 12 illustrates the types of trips that can be obtained from the
license plate follow-up survey technique. Like the license plate matching technique, the
lighter-shaded arrows represent the trips from one entry node to all other exit nodes (E-E
trips). However, instead of one dark-shaded arrow that aggregates all E-I trips from the
external station to the internal TAZs, information provided from the license plate follow-
up survey technique provides information on the distribution of the E-I trips to each of the
TAZs inside the cordon line.
 31

Figure 12: Types of Trips from License Plate Follow-Up Survey Technique

License plates must be recorded roadside using one of the methods described
above in the license plate matching section. With this method, however, it is important
to record the full license plate of the vehicle in order to contact the proper state motor
vehicle department (the Indiana Bureau of Motor Vehicles does not have information on
out-of-state license plates) and a partial license plate will not be useful. The advantage
over license plate matching is that obtaining each and every passing license plate is not
as critical because the plates are not being matched to another observation station,
although it is still important to record as many license plates as possible. The license
plates will likely have to be transcribed into an acceptable format and sent to each of the
respective departments of motor vehicles to obtain addresses.
 32

Various DMVs may take longer than others to respond, and it may take a long
time to get a response from all DMVs. It may be helpful to contact the DMVs in the
region ahead of the study. Addresses are usually obtained from the DMV on a cost-per-
plate basis, which may become quite expensive for large studies. In some states, the
DMV may refuse to provide addresses due to privacy issues. For legal information on
license plate follow-up surveys in Indiana, see Chapter 5.
In addition, some of the license plates may have been recorded or transcribed
incorrectly which will not provide any information. If license plates strings are recorded
without using video or photography, the state of the license plate may be difficult to
obtain (because the state name is much smaller than the serial number), especially on
commercial vehicle license plates, which often have single-color (often white)
backgrounds with black or blue characters. Unfortunately, out-of-state vehicles will likely
have different travel patterns than in-state vehicles (generally, a higher proportion of out-
of-state vehicles will be through trips).
Once the addresses are obtained, surveys are usually mailed out to the owners.
It is best if this can be done in three to five days, so the trip that is referred to in the
survey is still fresh in the mind of the driver. However, it may be that the owner of the
vehicle (especially commercial vehicles) was not the driver at the time the license plate
was recorded, or the owner may not want to respond to the survey for some reason.
Some people may be upset knowing that they were being watched and will not respond
due to privacy concerns.
Once the surveys are mailed out, low response rates are usually expected
(generally, 15 – 30%). However, providing more than one line of communication (such
as internet) may help increase the response. Also, using an internet response may
seem easiest, but by doing so, the responding sample will likely be small and
unrepresentative of the population of drivers observed on the road. There are many
ways the survey responses can become biased. Unlike the license plate matching
technique, a lot of time has to be spent reducing returned survey forms. In addition,
more money is spent for printing and mailing the survey forms (Quiroga, 2000).

2.4 Vehicle Intercept Survey Technique

Unlike the two described above, this technique requires interaction between the
driver and observer on the roadway. To conduct an OD study using this technique,
 33

stations are selected where trip information is desired. All vehicles or a random sample
of vehicles are then stopped along the roadway where drivers will voluntarily undergo a
roadside interview or be provided with a survey (to be completed after their trip and
mailed back). Both methods may be used in the same study (generally when roadside
interviews cause backups along the roadway upstream from the interview station), but
both methods should not be used on the same driver. Drivers should always be made
aware of the upcoming study via warning signs and traffic control barriers, and may be
intercepted with the help of police if necessary. Furthermore, an announcement to the
general public via radio, television, or newspaper may be appropriate. In some states,
this technique may be illegal. For legal information regarding vehicle intercept surveys
in Indiana, see Chapter 5. Figure 13 illustrates the typical layout of a vehicle intercept
station.

Figure 13: Typical Layout of a Vehicle Intercept Station

Source: Robertson, Manual of Transportation Engineering Studies (1994)

This technique is advantageous because it typically yields more trip data than a
license plate match. The questions are also adaptable, depending on the circumstances
of the study (Virkud, 1995). The response rates are much higher than a license plate
 34

follow-up survey described in the previous section, and are much less likely to be
biased.
This technique is more intrusive to the driver than the license plate matching
technique. In addition, vehicles are stopped on the roadway where they interact with
interviewers at a station. This can cause delays for the drivers and other traffic and pose
a safety hazard for the staff. This technique also requires a lot of manpower in planning
and conducting the OD study.

2.4.1 Roadside Interviews

Roadside interview is one method of the vehicle intercept technique. Once
vehicles are safely intercepted (via a flagger or policeman) and removed from the
flowing traffic lanes, the observer conducts an interviewer from the driver’s side of the
vehicle. This interview should last approximately 30 seconds and not exceed one
minute. The interview should seek information about the type of driver, vehicle, and trip.
Some vehicle information, such as classification and auto occupancy, may be obtained
before or after the interview. The interviewer may use a clipboard or some electronic
means (such as a laptop or PDA) to record information. In addition, a map may assist
the driver in answering some of the information about the trip. It is important, however,
that the map is legible. If the driver is unwilling or reluctant to provide any information,
he or she must be let go. For studies in which a sample of drivers is interviewed, a
proportional share of vehicle types should be sampled in order to prevent bias.
This type of vehicle intercept survey (VIS) generally yields a great deal of
information about one or several specific corridors. Information is collected real-time
directly from the person who is driving the vehicle. Most drivers are happy to comply,
and driver participation rates are usually above 90%. In addition, a laptop or PDA or
other electronic device can be used to store interviews and geocode origins and
destinations (Quiroga, 2000).
This method will typically require more manpower than any other technique.
Because the observers, supervisors, policemen, and flagmen are interacting with
vehicles on the roadway, it is inherently more dangerous than other roadside techniques.
In several states, such as Indiana, roadside interviews (RSIs) have not been
used in recent years, primarily due to motorist complaints. However, other factors, such
 35

as personnel safety, staffing requirements, travel delays, and privacy issues are
sometimes cited as reasons for not conducting RSIs.
This method is not suitable for all roadway sections. Traffic volumes on high-
level roadways such as freeways, expressways, and some arterials present a safety
hazard to both drivers and interviewers. If, however, rest areas or other easily
accessible roadside areas exist in which interviews can be conducted, this technique
may be used. The roadside interview is more appropriate in rural areas, particularly
along rural two-lane roads. Wherever the interview is conducted, adequate plans for
warning signs and traffic control barriers should be drawn up for proper field setup (see
Figure 13).

2.4.2 Postcard Questionnaires

This method can be used on its own or as a complement to the roadside
interview. The handout survey generally asks the same questions as conducted in the
roadside interview, however, the drivers voluntarily fill out the information and send the
survey back to the agency conducting the interview. The survey should be short enough
to fit on a postcard, and return postage should be prepaid. In order to track the types of
drivers responding to the survey, color-coded and/or numbered survey forms may be
used for each group (vehicle class, out-of-state drivers) in order to expand the data to
the entire population.
This method can be used on roadways with higher traffic volumes because they
require less interaction time with the driver. Drivers may not have to be directed off the
roadway; rather, postcards can be quickly handed out in the traffic lanes to every vehicle
as they stop at the roadside station. Like the roadside interview, adequate advanced
warning signs and traffic control must be in place for the safety of the drivers and
observers.
To complement roadside interviews, this method may be used when backups
occur upstream from the interview site. In this case, postcards may be handed to the
drivers and they are then permitted to leave. This will eliminate or reduce the delay and
number of angry drivers who are stopped to take the interview.
This method can be used as an alternative or in combination with roadside
interviews (RSIs). In this method, the same information is generally collected as in an
RSI, but the survey is conducted via a postcard that is handed to the driver, completed
 36

after the trip, and mailed back. A given number of personnel could hand out more
questionnaires than conduct roadside interviews.
The problem with this method is the lower response rate than with a roadside
interview. In addition, more of the questions may be skipped or answered incorrectly.
Generally, response rates for this method are between 15% and 30%. Furthermore, a
lot of time has to be spent reducing returned survey forms, and more money is spent for
printing them (Quiroga, 2000). There may be a bias in this type of survey, if non-
respondents (such as certain vehicle types or income levels) have different travel
characteristics and demographics than respondents. For example, surveys may not be
completed for several reasons: refusal to accept survey, failure to read it, failure to
understand it, failure to complete it, and failure to send it back (Bonsall, 1993).

2.4.3 Tag-on-Vehicle Survey

In this method, drivers are stopped at roadside stations where a color-coded
identifier is placed on the bumper, front window, or radio antenna of passing vehicles.
Each roadside station has one unique color assigned to it. Data collectors at each
station then record the passing vehicles’ tag color (if it has one) to determine the
percentage of vehicles coming from another station. Drivers are instructed to remove
the identifier at their next destination. With this method, a time stamp will not likely be
obtained.
The tag-on-vehicle method is a combination of the VIS and matching techniques.
Because the vehicles have to be stopped on the roadway in order for a tag to be placed
on their vehicle, it is considered a VIS. However, the tags are monitored as they pass
observers through subsequent stations on their trip, so it is also a type of matching
technique. The advantages of the tag-on-vehicle method are that it is quicker to conduct
than an RSI and easier to match between stations than license plates. However, time
stamps may not be collected, unless the vehicle is stopped again at the second station
to obtain that information.
On the downside, some motorists may not like the idea of physically attaching a
tag to their vehicle, and may disapprove of its placement or remove it before their
destination. Still other motorists may leave it attached even after they arrive at their
destination, which may cause a significant number of false matches if the vehicle is
 37

spotted later in the study. In addition, litter could become a problem if tags are not
secured or drivers do not dispose of them properly.

2.4.4 Summary of Methods for the Vehicle Intercept Survey Technique

Table 2 below lists some of the characteristics of the vehicle intercept survey
technique. Each column represents one of the three methods for conducting VIS
described in this chapter. The table entries qualitatively assess these characteristics so
that each method can be relatively compared with the other methods.

Table 2: Summary of Characteristics for Vehicle Intercept Survey Methods

RSI PC-Q TOV

Equipment costs high med low
Observer training high med low
Setup difficulty high high low
Subject to weather med med low
Power required yes no no
Require separate traffic counts yes yes yes
Travel Times no no no
*Note: RSI: Roadside Interview; PC-Q: Postcard Questionnaire; TOV: Tag-on-
Vehicle

2.5 Vehicle Tracing Technique

Vehicle tracing is a technique that utilizes some newer technologies to trace
vehicles unobtrusively through a study area. Theoretically, it could be utilized for
household travel surveys or the roadside station technique because it would collect
travel data on all vehicle trips within a study area including internal-external, external-
internal, and external-external (through) trips. Basically, it obtains the coordinates of
vehicles within the study area at regular time intervals. This information is obtained from
vehicles equipped with GPS navigation systems or cell phone users. Not only could
origin and destination information be obtained, but vehicle paths as well. There are,
however, many practical issues that have to be resolved in order to take advantage of
this technique.
 38

This technique is similar to the travel diary technique, but instead of recruiting
participants, training them, and distributing sometimes expensive equipment, the data is
recorded via technology and infrastructure that is already owned and used by the
general public such as cellular phones and GPS systems (such as the On-Star system in
GM vehicles). This information could then be obtained from a large amount of people
over long periods of time.
This technique is a relatively new technique in that it has only been used in a few
small applications, usually to monitor traffic speeds. The biggest disadvantage to this
technique is that the much of the public sees it as an invasion of privacy. As a
consequence, many of the private companies that provide this service are unwilling to
share this information.
In October 2005, the Missouri Department of Transportation contracted with
Delcan Corporation and an unnamed wireless carrier to provide real-time, statewide,
anonymous cell phone data for monitoring traffic speeds on 5,500 miles of roads in the
state of Missouri. Similar, but smaller projects are also underway in Baltimore,
Maryland, Norfolk, Virginia, and Atlanta, Georgia (Yahoo! News, October 2005).

2.5.1 Global Positioning System (GPS) Tracing

GPS was originally developed by the Department of Defense for military
purposes. There are 24 satellites orbiting approximately 11,000 mi above the earth’s
surface. These satellites can be used to measure location and provide navigational
information (driving directions) and time at almost any location on the surface of the
earth (Trimble, 2005).
In this method, a GPS-equipped vehicle (such as those with GM’s On-Star)
records latitude and longitude coordinates at a certain interval throughout its entire trip.
This information is then obtained real-time or post-trip (which is similar to the household
travel diary method). The data can then be analyzed to develop trip rates, average trip
lengths, origins and destinations, etc. for many household types and demographics.
In-vehicle GPS systems such as On-Star were installed for two primary reasons:
1) safety and security such as location tracking for emergency response, roadside
assistance, and vehicle theft and 2) navigation and convenience such as driving
directions and location-based information and convenience services (OnStar, 2005).
The advantage to being able to use this type of system is that it is essentially a GPS-
 39

assisted travel diary without the need to distribute the GPS equipment. In addition, for
each trip, not only are the origins and destinations recorded, but the actual path used by
the driver is stored (unlike the clipboard and PDA-assisted travel diary methods).
The disadvantage is that most customers would not approve of being tracked
without their knowledge, and it is not known how many would allow their data to be used
in an OD study. Furthermore, some information that is obtained via a formal travel diary
such as trip purpose and auto occupancy, is not recorded for each trip. Further
complicating this matter is the fact that only a small percentage of vehicles are equipped
with GPS, and those that do are likely to be owned by households with a higher-than-
average annual income. Therefore, the sample of vehicles being traced is not likely to
be representative sample of the total population.

2.5.2 Wireless Phone Tracing

The Federal Communications Commission (FCC) is currently requiring all
wireless phone providers to be able to provide location information on all its customers
for emergency purposes as part of the E-911 program. The program is supposed to be
completed by December 31, 2005, but is running behind schedule. The ultimate goal is
to automatically provide emergency dispatchers the location of the wireless phone on
which the call is being made within 50 – 300 meters (Federal Communications
Commission, 2005).
There are two ways wireless phone providers will obtain location information.
However, the end results in either case are the coordinates of the wireless phone. One
method is handset-based. This method uses a GPS receiver in the wireless phone,
which determines location and sends it to the provider. The second method, called
network-based, triangulates the wireless phone signal from multiple cell towers to
pinpoint the location of the wireless phone (Der Wann, 2002), either by measuring the
angle of the wireless signal (angle of arrival - AOA) or the difference in time the signal
reaches the various towers (time difference of arrival – TDOA). Currently, both methods
are used in the US. Of the major US wireless carriers, Verizon and Sprint use the
handset-based method, while Cingular and T-Mobile use the network-based method.
Like the GPS probe vehicle method, the location information obtained can then be
plotted to a GIS package to reveal travel patterns of persons carrying cell phones.
 40

Figures 14 & 15 illustrate the difference between the handset-based and network based
location determination technologies.

Figure 14: Handset-Based Location Determination using GPS

Note: PSAP – Public Safety Answering Point

Figure 15: Network-Based Location Determination using TDOA

Source: www.911dispatch.com
 41

The wireless phone tracing method has many of the same advantages and
disadvantages as the GPS method described in the previous section. The biggest
advantage to cellular tracing compared to GPS is that there is a much higher market
penetration. At the end of 2004, there were 182 million cell phone subscribers in the US,
up 23.4 million from 2003 (Cellular Telecommunications and Internet Association, 2005).
Cell phone ownership is likely to be more equally spread among the different cohorts of
age, income, and household size than GPS-equipped vehicles. Like GPS tracing, cell
phone tracing can also provide the actual routes of moving people rather than just the
origin and destination as in a clipboard travel diary or license plate match.
There are some disadvantages with this method. Information such as trip
purpose, auto occupancy, and possibly travel mode may be unobtainable. In addition,
most cell phone owners, like GPS-equipped vehicle owners, would not want to be traced
without their knowledge. However, due to the deep market penetration of cell phones, it
may be likely that there are enough volunteers to complete a study like the travel diary
technique. In addition, while the FCC is trying to implement Phase II of the E-911
program, it has not been completed in every state and county. Furthermore, the
accuracy of the locator points is not as accurate as with GPS, and depending on the
technology used by the wireless carrier, location information may not be provided for
some phones located in vehicles (because they need a clear view of satellites). Finally,
by tracing cell phones, person-trips, not vehicle trips are being recorded. This may or
may not be advantageous. For vehicle trips, it is possible that there could be multiple
cell phones in the same vehicle (such as a bus). Likewise, phones not in vehicles (for
example, pedestrians on sidewalks) will have to be filtered from those that are. On the
other hand, all travel modes could be traced, which would aid in developing activity-
based travel demand models.

2.5.3 Summary of Methods for the Vehicle Tracing Technique

Table 3 below lists some of the characteristics of the vehicle tracing technique.
Each column represents one of the three methods for conducting VIS described in this
chapter. The table entries qualitatively assess these characteristics so that each method
can be relatively compared with the other methods.
 42

Table 3: Summary of Characteristics for Vehicle Tracing Methods

GPS Cell
Equipment costs high high
Implementation Time high high
Subject to weather no no
Require separate traffic counts yes yes
Travel Times yes yes
Vehicle Paths yes yes

2.6 Mathematical OD Estimation Techniques

Over the years, a lot of research has been conducted on developing origin-
destination matrices using mathematical techniques to avoid the often time-consuming
and costly attributes of conducting a formal OD study, especially in large urban areas.
Most of these techniques rely on link traffic counts and some knowledge of a prior
matrix, usually one from a previous study. However, the number and location of traffic
counts, the quality of the prior OD matrix (in many cases, there isn’t one) often influence
the quality of the newly estimated OD matrix. In addition, many have additional
constraints that make them valid under certain assumptions (e.g., simple, uncongested
networks).
While there are many techniques available, no single estimation technique has
emerged that is widely accepted in practice. Therefore, these estimation techniques
were not evaluated as part of this project.

2.7 Summary of All OD Techniques

Each of the techniques and methods for conducting a Roadside Station OD
Study, the characteristics of each, and their advantages and disadvantages have been
described in detail in this chapter. In order to better compare those aspects, Table 4
summarizes the characteristics of the Roadside Station OD Study and qualitatively
compares each of the techniques.
 43

Table 4: Summary & Comparison of Roadside Station OD Study Techniques

 44

CHAPTER 3 – STATE DOT SURVEYS ON OD TECHNIQUES AND METHODS

To determine the extent of the use of the different types of OD techniques and
methods by other state DOTs, two surveys were prepared and sent to representatives
from each of the state DOTs. The surveys and the results are described below.

3.1 Survey on Roadside Interviews

In November 2004, a short survey was sent to the Highway Performance
Monitoring System (HPMS) coordinators at each State DOT to determine the usage and
legalities of roadside interviews (RSIs) in each state. The questions asked were: “Is
your DOT permitted to stop traffic to ask motorists questions such as their origins and
destinations? If so, to what extent do you use this travel survey method?” The HPMS
coordinator was asked to forward the question on to the appropriate person if he/she
was not able to provide the answer.
14 of the 50 states responded. All of the states responding said roadside
interviews were legal in their state, but only seven of the states said they are used (and
rarely used in three of those states). State DOTs that do conduct RSIs include
Delaware, South Dakota, Michigan, New Jersey, Ohio, Mississippi, and Utah. State
DOTs that do not conduct RSIs include New Hampshire, Wyoming, Connecticut,
Pennsylvania, Colorado, Rhode Island, and New Mexico (although consultants may
conduct RSIs in the New Mexico from time to time).
Of the states that do conduct RSIs regularly, most do with the assistance of the
police department. In Florida, however, it was the police department that decided to
stop participating in RSIs as a result of motorist complaints (Florida Highway Patrol,
2005).
Some states, like Delaware and Utah, prefer the postcard questionnaire method
over RSIs. While Delaware has conducted RSIs on freeway mainlines, most now are
 45

done via postcard during red phases on arterial streets. In New Mexico, RSIs can only
be conducted in rest areas.
The Michigan DOT, however, has revived the use of RSIs in the state with the
approval of the Attorney General’s Office and the Traffic and Safety Division, and has
used them frequently in the past two years. In 2004, 19 RSIs were conducted,
interviewing 25,000 drivers and generating only two dozen complaints. Michigan has
used this method on both two-lane and multi-lane highways with ADTs less than 30,000
vehicles. They have stopped motorists near toll facilities, rest areas, and on the
mainline. Postcard surveys were also conducted in some locations with a 25% response
rate.
Even though RSIs are legal in all of the states surveyed, most DOTs rarely or
never perform them.

3.2 Survey on all OD Techniques and Methods

To obtain more data regarding all the techniques, a survey of the 50 State DOTs
was conducted in April 2005 to determine the state-of-the-practice regarding policies and
use of data collection methods for origin-destination studies. The survey was distributed
to state DOTS via a contact list provided by the American Association of State Highway
and Transportation Officials (AASHTO). The survey is shown in Appendix 1 as it was
sent.
The initial deadline was set about six weeks after the survey was distributed.
The survey was sent again in its same form after the initial deadline to encourage more
response. Ultimately, a total of 19 states responded to the survey.

3.2.1 Results
• Does your state DOT have any written guidelines for conducting origin-destination
studies (for cities and towns located outside the jurisdiction of an MPO)? If so, how
could I obtain a copy of those guidelines?
Four out of 17 states (24%) utilized some form of written guidelines, ranging from
past experiences to TMIP documents.
 46

• Does your state DOT have a preferred data collection method for conducting origin-
destination studies (whether completed in-house or by a consultant)? If so, what is
that method?
Four out of 17 states (24%) have preferred method (one state indicated two) for
collecting OD data. Of the four states:
• Three indicated VIS
• One indicated LP matching with video
• One indicated LP follow-up surveys
• If your state DOT or its consultants have conducted any origin-destination studies in
the last five years, please complete sections A – F below. If not, skip to question 6.
Twelve of the 19 states (63%) indicated they have conducted at least one OD
study in the last 5 years. The following list indicates how many states utilized
each technique and method.
A. Roadside License Plate MatchingTechnique
Six of the twelve states (50%) used this technique:
• One state used the clipboard method
• Two states used the audio method
• One state used the laptop method
• Three states used the video method
B. Roadside License Plate Follow-Up Survey Technique
Four of the twelve states (33%) used this technique:
• All four states used the video method (three with manual
transcription, one with automatic transcription)
• All four states contacted vehicle owners by mail
C. Vehicle Intercept Survey
Eleven of the twelve states (92%) used this technique:
• Ten states used the roadside interview method
• Seven states used the postcard handout/mail-in method
D. Travel Diary
Eight of the twelve states (67%) used this technique:
• All eight states used paper travel diaries (two were part of the
2001 NHTS)
 47

E. Recall Interview
Two of the twelve states (17%) used this technique:
• Two states contacted households via telephone
• One state contacted households via mail
F. Vehicle Tracing
One of the twelve states (8%) used this technique. The method was
neither GPS nor cell phone tracing (a special study was created in which
random vehicles were followed to their destination).

• Is there any particular data collection method (not limited to those listed above) your
state DOT has utilized that has met or exceeded your expectations in terms of time,
cost, accuracy, etc? Please explain.
Four of the 19 states (21%) indicated one method exceeded their expectations:
• Three states indicated vehicle intercept surveys
• One state indicated LP matching with video
• Are there any methods that have failed to meet your expectations? Please explain.
Four of the 19 states (21%) indicated one method failed their expectations:
• Two states indicated LP matching with video
• One state indicated LP follow-up surveys
• One state indicated roadside interviews.

3.2.2 Conclusions
Several conclusions can be drawn from this survey. First, most states do not
have any sort of guidelines to refer to when conducting origin-destination studies.
Several of the responses indicated they were interested in obtaining any guidelines that
result from this study. Secondly, a wide variety of techniques have been used by the
responding states, with the most common to being vehicle intercept surveys, which
seems to indicate that, if conducted correctly, the VIS provides good OD information.
Finally, techniques that exceeded one state DOT’s expectations failed another’s (vehicle
intercept surveys and license plate matching with video were mentioned in each of these
questions). This may mean that certain techniques were conducted in situations that
provided poor results.
 48

CHAPTER 4 – LEGAL ISSUES OF VEHICLE INTERCEPT AND LICENSE PLATE

FOLLOW-UP SURVEYS IN INDIANA

Vehicle intercept surveys have not been conducted in the state of Indiana since
1991 due to an incident on an Indiana freeway that prompted a motorist complaint.
While the Indiana Attorney General intervened, it was unknown by INDOT officials if VIS
were officially made illegal or if the intervention applied to this single incident.
Likewise, license plate follow-up surveys have not been conducted in Indiana in
recent years. Critics of this technique often state that obtaining the vehicle owner’s
address from the motor vehicle bureau is an invasion of privacy. During this study, a call
to the Indiana BMV was made to inquire about obtaining information from a license plate
survey for OD purposes. The BMV responded that under no circumstances would
personal information (including owner addresses) be provided, even to other Indiana
government agencies.
To answer these questions, the Indiana Attorney General is being contacted with
assistance from the INDOT legal department for a clear and final ruling on the current
and future status of vehicle intercept and license plate follow-up surveys in Indiana. This
information will be provided under separate cover.
 49

CHAPTER 5 – EVALUATION OF EQUIPMENT FOR LICENSE PLATE DATA

COLLECTION

The purpose of this chapter is to evaluate the technology and equipment used in
various methods of recording license plate strings (either 4-character or full strings) from
a roadside station. These strings can then be matched with a list of license plate strings
obtained at other roadside stations (the license plate matching technique), or the strings
can be used to obtain addresses of vehicle owners to which a survey can be mailed
seeking specific trip information (the license plate follow-up survey technique).

5.1 Clipboard Technology & Equipment

For this method, very little equipment is required. Each observer needs a
clipboard with forms on which to record the license plate strings. Also required is a
device to keep time (clock, stopwatch, or wristwatch), which should be synchronized with
all other stations when conducting a study. A sufficient supply of writing utensils should
be provided as well.
Tablet PCs are an alternative to the standard clipboard if automatic transcription
of the license plates is desired. Tablet PCs may look much like a typical laptop;
however, the unit can be closed with the screen facing outward, which then becomes an
electronic clipboard. Handwritten information can be written directly on the screen using
a special writing pen. Tablet PCs have handwritten character recognition software that
converts handwritten text into digital characters in either real-time or upon command.
However, the accuracy of this software and its adaptability for roadside license plate
data collection is unknown, and it was not tested as part of this study. At this time, most
tablet PCs cost anywhere from $1000 to $2500 each.
 50

5.2 Audio Technology

There are basically two types of audio recorders: the traditional cassette tape
recorders and tape-free digital recorders. Both types were analyzed in this project to
determine if one type was better suited for license plate data collection.

5.2.1 Analog Cassette Audio Recorders

Most people are familiar with this type of audio recorder. The controls and
functions on it are simple and easy to understand. The recorder features two recording
modes: low speed and high speed. On the low-speed “normal” setting, the cassette tape
lasts as long as the time printed on the cassette itself. For example, a 60-minute
cassette can store 60 minutes of data (30 minutes per side). The high-speed setting
records a higher quality of sound, however, the cassette can only record for half the time
printed on the cassette. For example, a 60-minute cassette can store 30 minutes of data
(15 minutes per side).
Some audio cassette recorders have a voice-operated recording (VOR) feature,
also known as voice-activated recording. When the recorder senses no sound in the
room, the tape will automatically pause after a few moments, and resume once sound
returns. If the recorder is being used along the roadside, the background noise from
passing vehicles (especially on high volume roads) is usually enough to prevent the
VOR from pausing in the first place, which defeats the purpose for which it was intended.
Instead, if it is desired that long periods between vehicles are not recorded, the pause
button can still be used manually. However, it is important for the observer to remember
when the recorder is paused so that recording can begin again when the next vehicle
approaches. In addition, the time should also be updated during delays between
vehicles, usually no less than every one or two minutes.

5.2.2 Digital Audio Recorders

Digital recorders differ from cassette recorders for a number of reasons. First,
there is no external storage medium in a digital recorder. Rather, the data is stored
internally, usually in .wav files. The data is stored in files, so recordings are organized
much like songs on a music CD. The files can also be downloaded to a computer,
where the information can be digitally manipulated and stored.
 51

Like the cassette recorder, digital recorders typically have several recording
modes. Similarly, the higher the quality of the recording, the faster the storage is used.
The recorder may also have a microphone sensitivity adjustment. Usually, there
is a setting for dictation and another for recording sounds in all directions. For recording
license plates along the roadside, the microphone sensitivity should likely be set on the
dictation setting to minimize the amount of background noise that is recorded, which can
reduce the quality of the license plate records during playback.
Compared to the cassette recorder, the digital recorder is a little more
complicated than the cassette recorder to run at first, mainly to learn how the file storage
system works. Because there is no external storage medium like a cassette, the data
has to be deleted or downloaded when it becomes full. For a long OD study, the data
would have to be downloaded several times during the study period (which requires a
laptop and time away from continuously recording plates) or a multiple recorders at each
station.
Like the cassette recorder, digital recorders may also have the VOR feature,
although it may have a different name. For either recorder, it is recommended that this
feature be turned off.

5.3 Audio Equipment

Two handheld audio recorders, one analog and one digital, were purchased for
this study. The features of each are described below, followed by an evaluation of each
after being used to record license plates at a roadside setting. It should be noted,
however, that audio can also be recorded to other devices, such as a laptop computer,
which may be done if speech recognition software will be used to transcribe the license
plates in real time. Speech recognition technology was not evaluated as part of this
project. However, Washburn (1997) was able to obtain over 95% accuracy when
recording 525 vehicles per hour in field tests.

5.3.1 Sony Micro-cassette Audio Recorder

The first recorder, a Sony M560-V, is an analog micro-cassette recorder. It was
purchased for $33. The recorder features two recording speeds, the faster of which
produces higher-quality recording. It also has voice-operated recording (VOR), which is
used to save cassette space and battery power by automatically stopping recording
 52

when noise levels are minimal. The VOR sensitivity feature can be set on high, low, or
off. The recorder also contains a tape counter and one-finger rewind, fast-forward, and
pause for easier review during playback. The recorder runs on 2 AA alkaline batteries or
an AC adaptor (not included). The microphone is built in on the end of the unit (separate
from the playback speaker). This particular model does not have a separate microphone
jack, but it can be found on other Sony models. An earphone jack and earphones are
also included. The dimensions of the unit are 2 ½” x 4 ¾” x 1”, and it weighs 4.0 oz. A
4-pack of standard micro-cassettes with 60 min of recording time per cassette costs less
than $5.

5.3.2 Olympus Digital Audio Recorder

The second recorder is an Olympus VN-240PC digital voice recorder. It was
purchased for $69. Rather than record to cassette tapes, recordings are stored in .wav
files in four folders (A, B, C, S) for easier management. Each folder can hold 100 files.
Associated automatically with each file are the recording date and time, recording time
length, and audio quality. Files can be sorted, stored, and played back via the display
screen on the recorder.
The Olympus model has three recording modes: LP (lowest quality), SP, and HQ
(highest quality). The recorder can store a maximum of 245 minutes, 133 minutes, and
88 minutes, respectively, for each recording mode. There are also two levels of
microphone sensitivity: high – to be used to record sound from all directions, and low –
for dictation use. Like the VOR on the Sony model, this model contains variable control
voice actuator (VCVA), which can be turned on or off. Index marks (up to 10 per file)
can also be set while a file is being recorded. These can be used to find important times
in the file during playback. The recorder also contains microphone and earphone jacks,
although the accessories are not included.
This model also contains software to transfer audio files to a PC. A USB cable is
included. The Windows-based software can be used for storage and management of
recordings, playback, and direct recording to a PC. The recorder does not have a
conventional pause button. Instead, if recording is stopped and resumed, the resumed
portion is recorded into a new file. This model has a maximum of 100 files, regardless of
whether all the recording time has been used up. The dimensions of the unit are 3 ¾” x
1 ½” x ¾“, and it weighs just 2.4 oz.
 53

5.4 Evaluation of Audio Equipment

Because it is necessary to stand as close to the roadway as is safely possible in
order to see license plate information, there is generally a significant amount of
background noise due to the traffic itself. This noise also increases as the speed of the
vehicles increase. Therefore, it is important to make sure that, when using an audio
recorder to collect license plate data, the noise from the traffic does not prevent the data
on the audio recorder from being heard clearly during playback.
In order to compare the performance of the recorders, field testing was
conducted on the shoulder of a high-speed, four-lane highway. The results of the testing
of various features of the recorders were compared and are discussed below.

5.4.1 Audio Quality

As stated in the previous section, both recorders have different settings that
regulate the quality of the data recorded. For the cassette recorder, the faster the tapes
run, the higher the quality of the recording (and lower the storage capacity). Likewise,
while there is no “tape” in the digital recorder, higher quality is traded for lower capacity.
Tests were conducted in the same location on the roadside using two audio
recorders. License plate data was recorded on each of the settings and later reviewed
to determine if any of the data was indecipherable. The recorders were tested using the
manufacturers suggested settings for microphone sensitivity (if applicable), and each
recorder was placed the same distance (approximately 1 foot) from the observer. The
use of microphones or headsets were not evaluated on either type, rather, license plate
data was spoken directly into the microphones built into the recorders.
For both recorders, the highest quality audio setting is significantly better than the
lower settings. In situations with lots of background noise, the recording becomes
“scratchy” and reduces the clarity of the voice recording. Therefore, it is recommended
that the audio settings of the recorder be set on the highest quality to avoid loss of
license plate data.

5.4.2 Voice-Operated Recording

Both the recorders tested as part of this project have voice-operated recording.
The idea behind this feature is to preserve storage space by automatically stopping the
 54

recording when there is no sound present. On the roadside, it is unlikely, especially on

high-flow roads, that there will ever be a long enough silence to activate this feature. The
recorders were tested, however, to determine if this feature should be turned on or off
during a study.
For the cassette recorder, this feature should be turned off. This is because
recording resumes by an observer speaking into the recorder, the first part of the data
spoken into the recorder is lost. For example, if an observer speaks “A123” into the
recorder, only “123” is heard during playback.
The digital recorder, on the other hand, did not have the same problem as the
cassette recorder. However, it is unknown if all digital recorders are able to resume
recording without cutting off the first piece of information. Therefore, it is recommended
that VOR not be used on either type of audio recorder during an OD study. Because
audio recorders will likely be used in noisy situations, the benefit of VOR is probably
minimal.

5.4.3 Storage and Power

Because the high-quality audio settings are recommended to be used, the
storage capacity of each recorder is reduced. For the analog cassette recorder, the
micro-cassettes can record 15 minutes of data per side. This particular digital audio
recorder can record 88 minutes of data on the high-quality setting. The problem with the
digital recorder is that there is no external storage medium. The data can be recorded
directly to a laptop in real-time (this is beneficial if speech recognition software will be
used to automatically transcribe the data). Otherwise, the data must be downloaded
when all of the storage is used, which also requires a laptop in the vehicle. With the
analog cassette recorder, however, a new tape can be inserted relatively quickly.
Both recorders run on AA alkaline batteries. For long studies, the batteries may
need to be replaced. This task can be coordinated and planned among the observation
stations ahead of time, and can be completed in a relatively short amount of time.
While both recorders have advantages and disadvantages, the cassette recorder
is recommended for use in an actual OD study with manual transcription of the license
plate data. Compared to a digital recorder, the quality of the audio is similar, it is easier
to operate, and it has external storage that can be changed relatively quickly and easily.
 55

5.5 Laptop Technology & Equipment

The laptop method requires a laptop computer or some other electronic device in
which the license plate string can be keyed in using a standard keyboard. No
specialized software is required to store the license plate strings. Microsoft Office
software (Excel or Access) is likely to suffice. However, if more accurate time stamps
are desired for each license plate string, other software could be used. A simple data
entry form (for the license plate information) could be created in Access that contains a
macro that time stamps each license plate string with the current time as they are being
entered into the computer.
For this project, a Dell Latitude C600 was used. However, almost any type of
laptop could be used. For the purposes of this study, license plates were recorded in
Microsoft Excel, although another self-written program could be used to automatically
time stamp each license plate entry, store the data, and later, match the data.
There are a couple of important issues to consider when using a laptop on the
roadside. First, depending on the length of the study, the laptop may require additional
power beyond that of its regular battery. This may be achieved by using a higher
capacity battery or an adapter to connect to an automobile power source. Secondly, the
visibility of the screen may be reduced due to glare produced by the sun. This may
affect the quality of the license plate data being entered by an observer. Finally, most
laptops do not have a separate number keypad (as found on most desktop computers).
The amount of time it takes to enter a series of digits will likely be longer if a number
keypad is not available.

5.6 Digital Video Technology

A lot of license plate data collection has been conducted using analog
camcorders for traffic data collection. In recent years, however, digital video technology
has advanced well beyond VHS analog technology, and prices continue to fall. In 2005,
digital camcorders cost as little as $300. In addition, many manufacturers have stopped
producing analog cameras altogether. However, it is important that digital camcorders
can be easily utilized to produce adequate-quality video for use in OD studies. So what
exactly are the benefits and characteristics of digital video that make it more appealing?
The following sections discuss some of those characteristics of digital video and identify
 56

important features to look for when purchasing digital camcorders for the purpose of
recording license plates on the roadside.

5.6.1 Resolution
Resolution is one of the primary benefits of digital video. While 8mm analog
(VHS) camcorders have up to 250 lines of resolution, digital cameras have up to 520
lines, although the actual resolution is likely to be lower and varies from model to model.
Hi8, a higher-quality 8mm, can achieve only 400 lines. Furthermore, digital video quality
is not reduced by making copies or in storage over time, and editing is much easier.

5.6.2 Recording Media

Digital cameras currently have three types of recording media: tape, DVD, and
flash. MiniDV digital videotape is currently the most common. It is produced by almost
all manufacturers. (Sony, however, uses a digital tape format called Digital8.)
DVD camcorders have also become popular in recent years. These camcorders
record on smaller-sized DVDs in one of three formats: DVD-R, DVD-RW, or DVD-RAM.
A good feature of DVD recording is that editing can be done directly on the camera.
However, not all DVD players (especially older ones) can play the DVD-RW and DVD-
RAM formats, and recording time on one DVD is less than that of a MiniDV.
The other type of media is flash memory, such as SD or Compact Flash. Most
camcorders use this type for still digital pictures in addition to the tape or DVD media.
However, some camcorders record video to flash as well, although it hasn’t really caught
on, and storage capacity is much smaller than with MiniDV or DVD.

5.6.3 Playback & Editing

Digital video cannot be played back through a standard VCR like VHS tapes.
Instead, digital video can be played to a TV directly from a camcorder, or the video can
be downloaded to a computer for editing and storage. The best way to transfer video is
through the IEEE 1394 serial bus, more widely known as a FireWire (although it is also
called iLink by Sony). Most camcorders produced today have this port; however, not all
computers do, especially older models. A newer, high-speed USB port, USB 2.0, is also
capable of video transfer. Once the video is transferred to a computer, it can be edited
 57

and burned to DVDs for storage. If no editing is desired, it can always be stored on the
MiniDV tape itself. These currently cost about $5-7 per (one-hour) tape.
Some of the more technical details of a digital camcorder for recording vehicle
license plates for an OD study include color, magnification, focus, shutter speed, and
exposure. These are discussed in greater detail below.

5.6.4 Charge-Couple Device (CCD)

For the best color, the CCD size and number are important. CCD stands for
charge-couple device, or more simply, the “chip”. The CCD is the digital equivalent of
film. Its size is one of the biggest factors in the price of a digital camcorder. Sizes in
consumer-grade camcorders typically range between 1/6” to 1/3”, although professional-
grade cameras have up to 2/3” or 1” CCDs. CCD size is most important in low-light
situations. Larger CCDs also have more pixels, which allow the picture to be enlarged
without pixilation. However, above 340,000 pixels, the sharpness of the image improves
at a decreasing and negligible rate.
Another difference between camcorders is that it may contain one or three
CCDs. In three-CCD camcorders, there is one chip for each of the primary colors.
Camcorders with three CCDs have brighter and more lifelike colors than one-CCD
camcorders, and cannot presently be found on the market for less than $800.

5.6.5 Magnification (Zoom)

Camcorders typically have both optical and digital zoom. Optical zoom is an
actual movement of the lens of the camera and typically has magnification up to 25x.
Digital zoom, however, is just the enlargement of pixels, and will result in pixilation. Most
camcorders boast digital zooms of 500-1000x, but it is largely a marketing tool, as the
image will be completely illegible at such a high digital magnification.

5.6.6 Focus
Focus is an important part of recording video, especially for license plates.
Camcorders usually have automatic focus and manual focus options. For manual focus,
the best type of manual focus is a focus ring, which is usually located around the lens.
Some camcorders have a jog dial (similar to a sliding switch) or a focus button. Sony
 58

has a “spot focus” feature, where the user can use the touch-screen LCD viewfinder to
place the focus on a specific object.

5.6.7 Shutter Speed

Shutter speed is a measurement of the opening of the camcorder’s shutter in
cycles per second. Low shutter speeds admit more light to the CCD than high shutter
speeds and will result in a blurring of fast moving objects such as vehicles or athletes.
Too little light admittance at a high shutter speed can be compensated by adjusting the
exposure (or aperture). In some models, this adjustment is done automatically.

5.6.8 Exposure/Aperture
While the shutter controls the interval in which light is admitted, the exposure or
aperture controls the amount of light during any given interval. The bigger the aperture,
the more light is let in, which results in a brighter picture. Because most consumers (and
transportation professionals) are not photography experts, most camcorders produced
today have an automatic mode that controls the exposure. Many also have programmed
auto exposure (and shutter speed) modes such as “sports”, “candlelight”, “spotlight”,
“snow”, etc. (easycamcorders.com, 2005)

5.6.9 Other Features

Over the course of conducting this research with several different camcorder
models, there were a couple other features that were found to be helpful, but not
available on every model. The first, slow motion (and frame-by-frame) playback, was
available on only one model. This feature was very helpful for analyzing video of high-
speed vehicles that only appeared on the viewfinder for fractions of a second. For some
of the models, there is a slight delay when the video is paused. Because of this, it was
very difficult to actually stop the video when the license plate was in the field of view, in
which case, the video had to be rewound, replayed, and paused repeatedly until the
desired frame appears still on the screen. This process could take several tries, which
will result in longer analysis time and frustration for the observer. With slow-motion and
frame-by-frame capabilities, the observer is able to slow down the video as the front of
the vehicle entered the screen, then step forward (or backward) until the desired frame
appears in the screen. This camcorder model also had a remote control from which the
 59

video playback was controlled. If multiple camcorders are used in a study, not all need
to have slow-motion capabilities. Instead, the one camcorder with these features can be
used to play back the video recorded on all other camcorders.
Another feature that was only available on one of the camcorders was a time
stamp that was precise to the nearest second. While all models typically measure the
location of the tape in an hour, minute, second, and frame format, this information can
only be shown on the viewfinder, but not the television screen during playback. The
information that can be viewed on the television screen during playback is the actual
date and time. However, only one of the models measured the time to the nearest
second (the other two measured the nearest minute). While this precision is likely not
necessary for the average OD study, a time stamp to the nearest second may be useful,
for example, in conducting a travel time study on a corridor.

5.7 Digital Video Equipment

The following table summarizes the features that were available on the
camcorders used for this project. The three camcorder models used were 1) Panasonic
PV-GS19, 2) Canon ZR-80, and 3) Canon GL2.

Table 5: Summary of Camcorder Features

Panasonic
Feature PV-GS19 Canon ZR-80 Canon GL2
Storage Format MiniDV MiniDV MiniDV
CCD Size 1/6" 1/6" 3 @ 1/4"
Pixels (Absolute/Effective) 680k/340k 680k/340k 410k/380k
Optical Zoom 24x 18x 20x
Shutter Speed 1/60 - 1/8000 1/60 - 1/2000 1/8 - 1/15000
Auto Focus Yes Yes Yes
Manual Focus Yes Yes Yes
Time Stamp to sec Yes No No
Frame-by-Frame Playback No No Yes
Cost $349 $329 ~$2500

Both the PV-GS19 and ZR-80 are low-end consumer camcorders, while the GL2
is a professional grade camcorder. It can be seen that, while the GL2 is much more
expensive than its low-end counterparts, the only feature that is significantly different is
 60

the number and size of the CCDs. For license plate recording, this feature is not very
important, because the quality of the color on the video will not have much (if any) effect
on the clarity of the license plate digits.

5.8 Evaluation of Digital Video Equipment

The following field tests illustrate the issues to consider when using a camcorder
for license plate data collection. These issues relate to camcorder settings (such as
exposure, shutter speed, and magnification) to traffic conditions (such as vehicle speed
and traffic flow rates). These tests highlight the camcorder features to look for when
purchasing a digital camcorder for license plate data collection, as well as the optimal
camcorder settings for a specific set of traffic conditions.

5.8.1 Typical Roadside & Overhead Setup

Figures 16 and 17 illustrate the typical setup for all testing conducted with a
camcorder from a roadside perspective and an overhead perspective. The following
paragraphs explain the dimensions and variables in the figures in greater detail.

Figure 16: Plan View of Typical Camcorder Roadside Setup

 61

Figure 17: Elevation View of Typical Camcorder Overhead Setup

In the figures above, the subscript h refers to horizontal dimensions, and the
subscript v refers to vertical dimensions. The distance from the lane edge at which the
camcorder is set (on a tripod) is noted by n. This distance should be minimized, but the
safety of the observer (camcorder operator) is most important. Advanced warning
and/or traffic cones should be utilized to emphasize the observer’s presence, especially
on high speed and rural roads.
The shooting angle is the angle between the center of the field of view and some
reference line. The shooting angle is denoted by θ in Figures 16 and 17. This angle is
measured from a line parallel to the centerline of the driving lane. Therefore, when θ = 0
degrees, the camcorder is aimed parallel to the road; and when θ = 90 degrees, the
camcorder is aimed perpendicular to the road.
The angle-of-view is the angle between the edges of the picture and the
camcorder lens, and is denoted by Φ in Figures 16 and 17. For small zooms and focal
lengths, Φ is a wide angle. As the zoom and focal length increase, Φ decreases. The
relationship between zoom and angle-of-view will be discussed in greater detail later.
The shooting angle θ bisects the angle-of-view Φ.
The left (bottom) edge of the picture on the viewfinder is the left (bottom) edge of
the field-of-view. This line can be denoted as θa (in Figures 16 and 17) which is equal to
θ + ½ Φ. Likewise, the right (top) edge of the field-of-view, denoted as θb, is equal to θ -
½ Φ. As Φ decreases, θa and θb approach θ.
For a vehicle traveling on the roadway, its license plate will appear on the
viewfinder at point a, which is the point where the line along θa intersects with the
 62

centerline of the driving lane. Likewise, the license plate will disappear from the
viewfinder at point b, which is the point at which the line along θb intersects with the
centerline of the lane. The distance L between points a and b can be found using simple
trigonometry. The time the license plate remains in the field-of-view tL for a given Φ is a
function of the speed of the vehicle v. Values for each of these variables assume the
license plate is mounted on the center of the vehicle on a vehicle traveling down the
center of the lane.

5.8.2 Lighting Conditions

The legibility of license plates under various lighting conditions was evaluated.
All recording was conducted on the same roadway. The roadway on which the
recording was done has an east-west alignment and a speed limit of 35 mph. For each
lighting condition, several exposure settings were evaluated. All other camcorder
settings were held constant.
Most digital camcorders have preset exposure and shutter speed settings for
amateur videographers with point-and-shoot capabilities in mind. These include
automatic (which optimizes exposure for a given lighting condition), sports (for fast
moving objects), surf & ski (for high-glare situations), low-light (for poorly lit rooms), and
spotlight (for theatre-like conditions). This allows the videographer to avoid having to
find the optimal settings manually on his or her own.
Not all camcorders allow for manual adjustment of any or all of these features.
The Panasonic camcorder, however, that was used in this testing allows the user to
adjust all of them manually.
Four lighting conditions were tested: overcast, sun overhead, sun behind
camcorder, and sun in front of camcorder. Figure 18 illustrates the four lighting
conditions evaluated.
 63

Figure 18: Lighting Conditions Encountered on the Roadside

Upon evaluation of each of the settings, it is evident that lighting is not a problem
the vast majority of the time. In most conditions, the automatic exposure setting
provided an adequate amount of light that was neither too bright nor too dark to see the
contrast between the license plate string and the background of the license plate. There
are a few circumstances worth noting that are discussed in greater detail below.
Typically, the preset exposure settings had little visual effect on the quality of the
footage. The surf & ski mode (for high-glare situations) did not seem to have much
effect in sunny conditions. The automatic setting for exposure was relatively the same
as the preset modes under all lighting conditions.
Under the ‘sun overhead’ condition, many license plates were recorded that were
partially shadowed due to being inset into the vehicle. In all cases, the contrast between
the sun and shadowed parts of the plate did not affect the legibility of the license plate
string.
When the sun is behind the camcorder, glare (white-out of the license plate) is
very rarely a problem. Glare occurs when the angle at which the license plate is
mounted on the vehicle (uncontrollable) and the angle of the camcorder relative to the
sun (controllable) are equal. To avoid this, the camcorder angle should not be
positioned such that the sunlight hits the license plate and bounces directly towards the
camcorder. This, however, cannot be controlled for every passing vehicle. However, as
will be discussed later, the camcorder angle is also governed by traffic speed and
 64

vehicle flow so, ultimately, it is impossible to prevent glare in all circumstances for all
vehicles.
When the sun was in front of the camcorder, silhouetting of the vehicle
(appearing black in front of a bright background) was also generally not a problem,
however, in this test, the vehicle was not directly between the sun and the camcorder.
Obviously, the sun should not appear directly in the viewfinder, but this is also largely a
function of the alignment of the roadway relative to the location of the sunrise and
sunset, which varies by time of day and season.
It was not determined how much daylight had to be present in order for lighting
not to be a problem because the amount of daylight at a certain time of day changes
with the calendar. Generally, OD studies should not be conducted before sunrise or
after sunset for the safety of the observers. If possible then, studies using video should
be conducted during the season in which the start and end times occur during daylight
hours, preferably when the sun is already above the horizon.

5.8.3 Focus
Most digital camcorders have automatic focus. Some, however, include manual
focus as a feature. The camcorders used in this project all have automatic focus. In
most cases, automatic focus was not a problem. There are, however, a few
circumstances to avoid.
First, when choosing a location to set up the camcorder, select a location that
does not have any traffic signs, posts, poles, or other obstruction downstream. The
obstruction-free distance depends on the camcorder shooting angle and the distance at
which the camcorder is set up from lane edge. If a post or other object appears in the
viewfinder when shooting a vehicle beyond it, the camcorder may focus on the
obstruction in the foreground image and not the vehicle.
In addition, avoid shooting in an area where pedestrians can walk into the field-
of-view, for example, shooting across a sidewalk parallel to the roadway. This is
especially true when the shooting angle is less than 15 degrees, because the
pedestrians will remain in the field-of-view for a significant amount of time. If it cannot be
avoided, direct pedestrians around the field-of-view so they do not shift the focus from
the vehicles.
 65

5.8.4 Obstruction & Minimum Shooting Angles

When recording the license plate from a single vehicle traveling on a roadway,
the theoretical optimal horizontal and vertical shooting angle is zero. This, however,
means the camcorder is set up in the center of the traffic lane at a height of
approximately 3 feet (the height of the average license plate). In this circumstance, the
license plate would appear in the center of the camcorder’s field-of-view and remain
stationary while the license plate characters shrink as the vehicle travels farther and
farther away. Obviously, however, the camcorder can never be set up in the traffic lane.
The second-best location for camcorder set up off the roadway is either centered
above the traffic lane (e.g., on an overpass) or on the side of the road. If the camcorder
is set on an overpass, the horizontal angle is zero, but the vertical angle is not. In
addition, the vertical distance nv from the camcorder to the license plate cannot be less
than ~20 feet (the height of the overpass deck plus tripod height minus the height of the
license plate). Conversely, on the side of the road, the vertical angle is close to zero, but
the horizontal angle is not. In addition, the horizontal distance nh of the camcorder on
the side of the road cannot be less than ~7 feet (half the width of the traffic lane plus one
foot for the tripod). While any shooting angle can be achieved regardless of the
distances nv or nh, the zoom capabilities of the camcorder will also limit the minimum
shooting angle. This will be discussed in the next section.
From a vantage point on the roadside, a vehicle following another will eventually
(some distance downstream) block the view of the license plate of the first vehicle.
Given this following distance and the distance nh, an angle can be calculated. At this
angle, the license plate of the first vehicle will come into view just as the right-front
corner of the following vehicle enters the field-of-view. Likewise, in the overhead setup,
the top of the following vehicle enters the field-of-view just as the license plate of the first
vehicle enters the field-of-view. Figures 19 and 20 illustrate this. This obstruction angle
is a function of the distance between the rear of the first vehicle and the front of the
following vehicle (which usually increases as speed increases), the height of the
following vehicle (for overhead setup), the location of the license plate on the first
vehicle, and the lateral location of both vehicles in the traffic lane.
 66

Figure 19: Following-Vehicle Obstruction from the Roadside Perspective

Figure 20: Following-Vehicle Obstruction from the Overhead Perspective

Figures 21 and 22 illustrate the obstruction angle for three different following
times. Following time is similar to headway, except following time is measured from the
rear of one vehicle to the front of the next, and headway is measured between the fronts
of subsequent vehicles. For high speeds, the difference between following time and
headway is small. Following times are assumed to be constant for all speeds, but the
corresponding following distance increases as speed increases. As the following
distance increases, the obstruction angle decreases. Therefore, the obstruction angle
decreases as speed increases.
 67

Obstruction Angle θ'

from the Roadside Perspective
90.0
80.0 tf = 0.5 s

70.0 tf = 1.0 s
Obstruction Angle (deg)
60.0 tf = 2.0 s

50.0
40.0
30.0
20.0
10.0
0.0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75
Vehicle Speed (mph)

Figure 21: Obstruction Angle from Roadside Perspective

Obstruction Angle θ'

from the Overhead Perspective
90.0
80.0 tf = 0.5 s
Obstruction Angle (deg)

70.0 tf = 1.0 s

60.0 tf = 2.0 s

50.0
40.0
30.0
20.0
10.0
0.0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75
Vehicle Speed (mph)

Figure 22: Obstruction Angle from Overhead Perspective

In low, free-flow traffic conditions, most vehicles maintain a following distance of

at least 2 seconds. In this case, the angles corresponding to the 2-second following time
should be used. In high flow, urban conditions (especially downstream from traffic
 68

signals) or congested conditions, a greater percentage of vehicles have a following time

of less than 2 seconds. In these situations, the angles corresponding to the ½ or 1
second following times should be used. Regardless of the angle chosen, a small
percentage of vehicles will still be blocked from the camcorder’s field-of-view.
When setting up a camcorder, the minimum shooting angle will be equal to the
obstruction angle θ’ plus one-half the angle-of-view Φ. The angle-of-view will be
discussed in the next section.
When using camcorders, there may be some situations in which a curve in the
road may be used to the advantage of the camcorder. The goal of setting up on a curve
is that it may reduce or eliminate the shooting angle of the camcorder. This is
advantageous because it reduces the speed at which the license plate traverses the
screen and increases the amount of time the license plate remains present on the
screen, which is beneficial for manual transcription. Figure 23 illustrates an example of
this.

Figure 23: Setup on Curves in the Road

Depending on the radius of the curve and the distance between vehicles
traveling on the road (which increases with speed), the shooting angle may or may not
be reduced to zero. However, depending on the circumstances, that reduction may or
may not be advantageous.

5.8.5 Magnification and Angle-of-View Variations by Camcorder Model

Camcorder models are manufactured with a wide variety of lenses, which affects
the maximum optical zoom of the camcorder. An optical zoom works by physically
 69

adjusting the distance (called the focal length) between the lens and the CCD. Again,
the CCD is the digital version of film. Figure 24 is a simple illustration of a lens.

Figure 24: Dimensions of Typical Camcorder Lens

An optical zoom of 10x means that an object will be magnified ten times as much
as the same object at a zoom of 1x. The zoom is directly related to the range of focal
length of the lens. For example, if the focal length of the lens ranges from 2.1 mm to
50.4 mm, the lens has a maximum optical zoom of 24x (50.4/2.1). This means that an
object can be magnified up to 24 times. A focal length of 4.2 mm on that same lens
corresponds to a zoom of 2x, a focal length of 8.4 mm corresponds to 4x, etc.
Unfortunately, a zoom of Ax on one camcorder will magnify an object more or
less than a zoom of Ax on another camcorder. This is because the range of focal
lengths of the lens on the second camcorder can, and often will, be different. For
example, the lens mentioned previously had a focal length range of 2.1mm – 50.4 mm
for a maximum zoom of 24x (50.4/2.1). Another camcorder might have a focal length of
2.8mm – 56.0mm for a maximum zoom of 20x (56.0/2.8). If both of these camcorders
are set up side by side and focus on the same object at 1x (or any other magnification),
the object will always appear smaller through the first camcorder. The reason for this is
the angle-of-view.
The angle-of-view is directly related to the focal length. As the focal length
increases (and the zoom increases), the angle-of-view decreases. This enables the
magnified object to appear on the viewfinder or screen (which stays the same size, of
course). Theoretically then, at a zoom of 1x, the object shot by the first camcorder will
appear 75% (2.1/2.8) as large as the same object shot by the second camcorder. It
should be noted, that there is more than one angle-of-view, as illustrated in Figure 25.
 70

However, the ratios between the angles-of-view remain constant as the focal length
changes.

Figure 25: Illustration of Angles-of-View

Further complicating matters is that, for any single focal length, the angle-of-view
depends on the size of the CCD. CCDs come in many sizes, between 1” for
professional-grade models down to 1/6” or 1/8” for consumer-grade models. Larger
CCDs are necessary if the video will be viewed on very large screens.
The CCD is measured diagonally, much like television screens. Standard 35 mm
film, for example, actually measures 43.3 mm diagonally. It would be assumed then,
that a 1” CCD is 25.4 mm by simple unit conversion. Unfortunately, the lack of
standards complicates this even more, because 1” is just a nominal measurement. The
actual diagonal measurement of any size CCD is roughly (but not exactly) 2/3 of the
nominal measurement. A 1” CCD effectively measures 16.0 mm diagonally.
Now that the effective diagonal length of the CCD is known approximately, an
actual angle-of-view can be calculated. Figure 26 illustrates the relationship between
focal length, CCD size, and angle-of-view. Table 6 shows the actual angle-of-view for
the PV-GS19 and ZR-80 camcorders. By looking at the predicted versus the actual
values of the angles-of-view, a significant error still exists in that the predicted value is
consistently higher than the actual value. This may be because CCDs of the same
 71

nominal size may actually be significantly different (from model to model), or the actual
display of the zoom of the camcorder is not accurate in the viewfinder.

Horizontal Angles-of-View for Various CCD Sizes

180
35mm Film
160
1" CCD
140 1/2" CCD
Angle-of-View (deg)

1/4" CCD
120
1/6" CCD
100

80

60

40

20

0
0.0 10.0 20.0 30.0 40.0 50.0 60.0
Focal Length (mm)

Figure 26: Relationship between Focal Length & Angle-of-View by CCD Size

Table 6: Predicted vs. Actual Angles-of-View

Panasonic PV-GS19 (1/6" CCD)
Zoom FL Predicted Actual % Diff
1x 2.1 64 45 29.7%
2x 4.2 34 30 11.8%
4x 8.4 18 13 27.8%
8x 16.8 8.8 6.9 21.6%
16x 33.6 4.4 3.4 22.7%
24x 50.4 3.0 2.3 23.3%
Canon ZR-80 (1/6" CCD)
Zoom FL Predicted Actual % Diff
1x 2.8 50 33 34.0%
18x 50.4 3.0 2.2 26.7%

All of this boils down to the fact that, for shooting license plates at an angle of θ
at a distance of nh feet from the lane edge, an optimal zoom cannot be specified to
ensure that the license plate characters will not be too small or too large for transcription
during playback. As a result, the magnification of the license plate will have to be
 72

roughly specified relative to the size of the viewfinder on the camcorder. This will be
discussed in the next section.

5.8.6 Magnification
Because magnification differs among camcorder models, the amount of
magnification required for any setup location will be defined by the size of the license
plate in the center of the field-of-view relative to the field-of-view. The minimum and
maximum magnification will be discussed below in addition to the magnification
limitations on the minimum shooting angle.
For any camcorder set up on the roadside or overhead, magnification will likely
be necessary in order for the license plate characters to be large enough to be legible
during playback. On many camcorders, the quality of the picture as seen on the
viewfinder is generally lower than as it appears during playback on a television.
Therefore, it may be difficult to know in the field if the magnification is enough for the
license plate characters to be legible. While there is some leeway, a good rule of thumb
is to magnify until the rear of the vehicle fills the width of the field-of-view. The ratio of
the width of the license plate w to the width of the viewfinder W is approximately 0.15.
For large shooting angles (above 30°), this ratio may be reduced slightly to adjust for the
skewness of the shooting angle relative to the rear of the vehicle. Figure 27 illustrates
this magnification ratio.
 73

Figure 27: Required Magnification of License Plates

While the minimum magnification is necessary for legibility, magnification much

beyond 0.15 should be avoided. This is because, when magnified too much, license
plates that are mounted at different heights (especially on trucks) may not appear within
the field-of-view at all.
Finally, when the camcorder shooting angle is set, the optical magnification
capabilities of the camcorder may not allow the minimum ratio to be achieved. In
addition, digital magnification should never be used for a license plate survey. In this
case, the distance n from the camcorder to the traffic lane should be reduced. This
however, is not possible from an overhead perspective, and may be limited from a
roadside perspective due to safety issues. If the camcorder cannot be moved, then the
shooting angle will have to be increased until the magnification ratio can be achieved.

5.8.7 Optimal Shutter Speed

The shutter speed is a very important aspect in setting up the camcorder. A
higher shutter speed can record moving objects more clearly. At a low shutter speed,
the object will appear blurry or streak across the field-of-view. The optimal shutter speed
 74

depends upon the speed of the vehicle perpendicular to the line-of-sight of the
camcorder. Figure 28 illustrates this.

Figure 28: Vehicle Speed Perpendicular to the Camcorder Line-of-Sight

If the camcorder were set up in the center of the traffic lane at the height of the
license plate (shooting angle equals 0°), the vehicle would not move laterally across the
screen (although it would get smaller as it moved farther away). In this case, a very low
shutter speed would be acceptable. However, if the shooting angle of the camcorder
was 90°, the vehicle’s speed would appear laterally across the screen but not into the
screen. In this case, a high shutter speed is required to prevent streaking of the picture.
Because the shutter speed is sensitive to the lateral movement (or perpendicular
movement) of the vehicle, the perpendicular vehicle speed needs to be calculated in
order to find the optimal shutter speed. Table 7 illustrates the perpendicular vehicle
speed as a function of the actual vehicle speed and shooting angle.

Table 7: Perpendicular Vehicle Speed (mph)

Shooting Angle (Line-of-Sight Angle), degrees

0 5 15 30 45 90
5 0.0 0.4 1.3 2.5 3.54 5.00
Actual Veh. Speed (mph)

15 0.0 1.3 3.9 7.5 10.6 15.0

25 0.0 2.2 6.5 12.5 17.7 25.0
35 0.0 3.1 9.1 17.5 24.7 35.0
45 0.0 3.9 11.6 22.5 31.8 45.0
55 0.0 4.8 14.2 27.5 38.9 55.0
65 0.0 5.7 16.8 32.5 46.0 65.0
75 0.0 6.5 19.4 37.5 53.0 75.0
 75

Using the appropriate perpendicular vehicle speed, Figure 29 provides the

optimal shutter speed camcorder setting. These shutter speeds were determined by
manually adjusting the shooting angles and shutter speeds until the characters of
recorded license plates appeared clearly during playback.

Optimal Shutter Speed

10000
9000
8000
Shutter Speed (1/y)

7000
6000
5000
4000
3000
Optimal Observed
2000
Optimal Assumed
1000
Minimum
0
0 10 20 30 40 50 60 70
Vehicle Speed (mph),
Perpendicular to Camcorder Line-of-Sight

Figure 29: Optimal Shutter Speed

5.8.8 Left Lane Obstruction

Generally, it is not desirable to use one camcorder to record multiple lanes of
traffic, especially on high volume roads. The reason for this is because vehicles in the
lane closest to the camcorder will block the view of vehicles in the far lanes. This
percentage of the time the far lane is blocked depends upon the amount of traffic flow
and the shooting angle of the camcorder. However, for certain roadway conditions, such
as undivided multilane roads, one camcorder will have to record both lanes because it is
not possible to shoot the far lane unobstructed from either side of the road. Figure 30
illustrates the left-lane obstruction issue.
 76

Figure 30: Obstruction of License Plates in Left Lane

The percentage of the time that the left lane is obstructed increases as vehicle
speed decreases and camcorder shooting angle decreases. The graphs in Figure 31
illustrate these obstruction factors for four different vehicle speeds.

Left Lane Obstruction Factor Left Lane Obstruction Factor

Speed = 15 mph Speed = 35 mph

1.00 1.00

0.90 0.90

0.80 0.80

0.70 0.70
Obstruction Factor

Obstruction Factor

90 deg 90 deg
0.60 0.60
45 45
0.50 30 0.50 30
15 15
0.40 0.40
5 5
0.30 0.30

0.20 0.20

0.10 0.10

0.00 0.00
0 200 400 600 800 1,000 1,200 1,400 1,600 1,800 0 200 400 600 800 1,000 1,200 1,400 1,600 1,800
Right Lane Hourly Flow Right Lane Hourly Flow

Left Lane Obstruction Factor Left Lane Obstruction Factor

Speed = 55 mph Speed = 75 mph

1.00 1.00

0.90 0.90

0.80 0.80

0.70 0.70
Obstruction Factor
Obstruction Factor

90 deg 90 deg
0.60 0.60
45 45
0.50 30 0.50 30
15 15
0.40 0.40
5 5
0.30 0.30

0.20 0.20

0.10 0.10

0.00 0.00
0 200 400 600 800 1,000 1,200 1,400 1,600 1,800 0 200 400 600 800 1,000 1,200 1,400 1,600 1,800
Right Lane Hourly Flow Right Lane Hourly Flow

Figure 31: Left Lane License Plate Obstruction Factor

5.9 Photography Equipment

Photography equipment was not evaluated as part of this project. At this time,
there are few, if any, manufacturers that manufacture cameras that have the features
 77

and capabilities required for the purposes of conducting short-term license plate data
collection at temporary roadside locations. However, almost all of the issues related to
using video camcorders also apply to this method.
 78

CHAPTER 6 – DETERMINING LICENSE PLATE DATA COLLECTION METHOD

In order to evaluate which data collection methods are suitable for a specific set
of roadway characteristics and traffic conditions, license plate data was collected in the
field for both the license plate matching technique (which requires that 4 characters of
the license plate be recorded) and the license plate follow-up survey technique (which
requires the full string and state-of-issuance so the address of the vehicle owner can be
obtained through the appropriate DMV).
The process of recording a license plate string can be broken into two distinct
parts: 1) identifying the string, and 2) recording the string. Section 6.3 evaluates the first
step, while Section 6.4 discusses the second step of that process using the technology
and equipment discussed in Chapter 5.
Before evaluating the various methods for license plate data collection, however,
Section 6.1 defines the rules for 4-character and full string license plate data collection,
and Section 6.2 illustrates and discusses the current license plate variations used in
Indiana and surrounding states that may be encountered when conducting a license
plate survey.

6.1 License Plate Data Collection Rules

Because there are so many variations in the syntax and size of license plate
strings, several rules need to be created in order to ensure that the appropriate data is
being collected in the field. The following sections discuss some important issues for the
license plate matching and license plate follow-up survey techniques.

6.1.1 License Plate Matching Technique

In this technique, license plates are matched between observation stations. No
information is collected from the driver, either on the roadside or through a follow-up
 79

survey. Therefore, the only data that is required from each observation station is some
unique identifier for the vehicle. Recording four characters from the license plate will
virtually eliminate the possibility of a spurious match. This was discussed in Chapter 3.
The last four characters (not the first four) should be recorded. This is especially
necessary in Indiana, because the state uses county identifiers as the first two digits on
standard passenger vehicle license plates. Also, both letters and numbers should be
recorded. In the case that characters are stacked vertically, the top number should be
recorded first. Examples of various license plates will be shown in the Section 6.2 to
illustrate these rules.

6.1.2 License Plate Follow-Up Survey Technique

Like the matching technique, there is no interaction with the driver during the
observation of the vehicle. Instead, trip information is obtained through a mailed follow-
up survey. In order to mail the vehicle owner a survey, the full license plate has to be
recorded so the owner’s address can be obtained from the state motor vehicle
department. In order to contact the appropriate motor vehicle department, the state
must be recorded from the license plate. Therefore, all of the characters that appear
across the center of the license plate should be recorded. Other information that
appears on the license plate, such as vehicle type, is not necessary because the serial
number of the license plate is unique to that vehicle. For both techniques, the license
plates should be recorded from the rear of the vehicle. This is because Indiana issues
one license plate which is placed in the rear.
In many studies that utilize this technique, only the states that are geographically
adjacent to the state in which the OD study is being conducted are contacted for owner’s
addresses. In Indiana, these states include Michigan, Ohio, Kentucky, and Illinois.
Depending on the location of the OD study within the state, and the amount and type of
out-of-state vehicles using the roadways within the study area, the adjacent states may
be excluded (or other states included). If there are many out-of-state license plates
recorded, not sending a survey may bias the results of the study. The states that will be
included should be contacted prior to the OD study.
 80

6.2 License Plate Variations

License plates come in many different colors, fonts, and syntaxes. Generally,
each of the 50 US states has a standard style for passenger vehicles. However, there
are often special recognition plates and vanity plates that have different variations. Most
states have additional variations for commercial vehicles and government vehicles
(state, municipal, university, etc.). Fortunately, the dimensions of license plates do not
vary from state to state, but smaller license plates are often used for motorcycles and
small trailers.
While all states require license plates to appear on the rear of vehicles, some
vehicles do not have them at all (missing or temporary) while others have plates that are
damaged, dirty, blocked, or covered (by tinted plastic covers or frames).
The following paragraphs illustrate the current style of license plates issued in
Indiana, as well as some of the different variations found in other states. In addition,
some special cases are illustrated to highlight some of the potential problems that may
be encountered in the field when conducting license plate data collection.

6.2.1 Indiana Passenger Vehicle License Plates

The standard passenger vehicle license plate issued by the Indiana Bureau of
Motor Vehicles (BMV) is illustrated in Figure 32. The standard license plate has a one or
two-digit county identifier, followed by one letter, and finally, one, two, three, or four
numbers. There are currently two variations of this plate being used. The string on the
first plate shown has no spaces and bold characters which are all the same height. This
variation is being phased out and replaced by the second license plate shown. The
string on the second plate has the same syntax, however, the characters are less bold,
the letter is smaller in height, and spaces appear before and after the letter. The
information next to each license plate displays the color, syntax, and information that
should be recorded under the 4-character and full string rules.
 81

Figure 32: Indiana Passenger Vehicle Standard License Plates

Source of LP Images: Nicholson, 2005, *Syntax not shown in figure

In addition to the standard license plates for passenger vehicles, there are many
special license plates that recognize other various groups. According to the Indiana
Bureau of Motor Vehicles website, there are currently special recognition plates for 23
colleges and universities, and 34 for other organizations ranging from military,
occupational, and other non-profit groups. Figure 33 illustrates a few of these license
plates.

Figure 33: Indiana Passenger Vehicle Special Recognition License Plates

Source of LP Images: Indiana Bureau of Motor Vehicles, 2005
 82

The general format of the special recognition plates issued by the Indiana BMV is
the same for all organizations. All contain a unique logo on the left, followed by two
stacked letters that are also unique to the organization (the stacked letters “HT” will not
appear on any other special recognition plate besides the environment plate), followed
by one, two, three, or four digits.
Many other states also issue special recognition plates. They may or may not
have a similar format as Indiana. Therefore, if a license plate follow-up survey technique
is being conducted using the clipboard, audio, or laptop methods, it may be difficult to
obtain the state that issued that license plate, because it is less likely that the observer
will recognize the state issuing the license plate.

6.2.2 Indiana Commercial & Government Vehicle License Plates

In addition to the passenger vehicle license plates discussed above, other
license plate variations are issued to commercial and government vehicles. Figure 34
illustrates the different variations of these.

Figure 34: Indiana Commercial & Government Vehicle License Plates

Source of LP Images: Welby, 2005; Not all syntaxes shown for each vehicle type
 83

6.2.3 Out-of-State License Plates

Like Indiana, other states generally issue a standard passenger vehicle plate,
special recognition plates, commercial vehicle plates, and government-owned license
plates. While the height of the characters on the license plate is consistent, the font of
the characters and the size and location of the state’s name vary. In addition, the state
name is sometimes covered by license plate frames. While this is not a problem for the
license plate matching technique, it may become a problem for the license plate follow-
up survey technique. In some cases, such as passenger vehicle plates, the state can be
determined solely by the unique style of the plate. However, for commercial,
government, and other generic plates, this may not be so. Figures 35 and 36 illustrate
the formats of the standard passenger vehicle and commercial vehicle license plates for
states surrounding Indiana.

Figure 35: Out-of-State Standard Passenger Vehicle License Plates

Source of LP Images: Nicholson, 2005
 84

Figure 36: Out-of-State Commercial Vehicle License Plates

Source of LP Images: Welby, 2005

6.3 License Plate Identification in the Field

Identifying the license plate string is the first step in the process of recording it. It
is common to all of the methods. However, for the clipboard, audio, and laptop methods,
this step is completed on the roadside during the study. The process of identifying the
license plate string can be broken down into further sub-processes. To illustrate these
sub-processes more clearly, imagine standing on the side of the road as vehicles pass
by. In order to identify the license plate string of a passing vehicle, the following needs
 85

to be done: 1) locate the license plate on the back of the vehicle, 2) locate the first
character or fourth-to-last character on the license plate, and 3) identify (and memorize)
the license plate string (and state, if applicable). The reason the string has to be
memorized is that, depending on the speed of the vehicle and the recording method
being used, it is unlikely the license plate string will be visible at a second glance
(because the vehicle will have traveled beyond the limit of legibility).
The value of this identification time is unknown, but important for a number of
reasons (as will be discussed later in this chapter). Therefore, it is important that it be
measured.
The amount of time required to identify a license plate string is dependent upon a
number of things: the capabilities and condition of the person identifying the string (such
as age, familiarity, and fatigue), the location of the license plate on the vehicle, the type
of vehicle (passenger car, motorcycle, truck, trailer), the characteristics of the license
plate such as the contrast between background and digit colors, and characteristics of
the string such as font, color, syntax. In addition, the mix of vehicles on the roadway will
have some effect on the identification time, compared to an ideal (but unlikely) situation
in which all vehicles on the roadway have the same license plate style and syntax.
The string identification time is important for several reasons. First, the length of
time that a license plate is legible to an observer on the roadside should be greater than
the string identification time. In other words, vehicles should be traveling slowly enough
or the observer should have visual acuity that is strong enough to enable him or her to
see the license plate string legibly for at least as long as it will take him or her to identify
it. Figure 37 illustrates this distance.

Figure 37: License Plate Legibility Distance

 86

Secondly, in order to speed up the manual transcription process of license plate

strings recorded via video, it is desirable (although not required or even possible in some
cases) to have the license plate string be visible and legible in the field-of-view for at
least as long as the string identification time. By doing so, the video tapes can be
transcribed at normal speed rather than in slow motion or pausing on each vehicle.
Because the string identification time is unknown, it had to be measured. The
string identification time can be defined as the time that elapses from when an observer
begins to search for a license plate on the back of vehicle until the observer has
identified and is ready to record the string to some medium.
As an observer on the roadside trying to identify license plate strings, the search
for the plate does not begin until the vehicle is some distance downstream from where
the observer is standing. This is because the license plate is easier to read when it is
moving along our line-of-sight. As the vehicle is just passing the observer, the license
plate is moving almost perpendicular to the observer’s line-of-sight, which makes it very
difficult to read, especially at higher vehicle speeds. In addition, the license plate is not
directly facing the observer at this point (because the observer is on the side of the
road). Therefore, it is assumed that, if the observer is standing six feet from the lane
edge, he or she begins to search for the license plate when the vehicle is approximately
20-25 feet downstream from where the observer is standing.
The string identification time ends when the observer has identified and
memorized the license plate string. The observer has to memorize the string before
recording because license plate string may not be legible by the time the observer looks
a second time, usually because the license plate will be too far away and beyond the
limit of the observer’s visual acuity.
It was determined that the most accurate way to measure this string identification
time was to have the researcher use a self-operated stopwatch. The reason the times
were self-measured is because the researcher is the most accurate judge at which the
string identification time begins and ends. As a vehicle passed, the researcher began
the stopwatch as he or she began to look at the back of the vehicle. The researcher
stops the time just as the last piece of information on the license plate is identified. The
time was recorded, and the procedure repeated 50 times. These measurements were
taken on roadways with three different speeds (15 mph, 35 mph, and 55 mph) from a
location six feet from the adjacent lane edge with a line of sight of approximately 30
 87

degrees from a line parallel to the road. The results of this data collection are presented
in sections 6.3.2 and 6.3.3.
Even if all vehicles were traveling at a speed slow enough and the observer has
a visual acuity that is strong enough, some vehicles will be unidentifiable because of
missing, damaged, dirty, covered, or blocked license plates. These vehicles are not
considered in the distribution of identification times. For any given roadway, a small
percentage of vehicles will be unidentifiable which, if unknown, introduces error when
expanding the data. In this case, if a vehicle is unidentifiable at one observation station
it is also unidentifiable at the next observation station. This is unlike another type of
error, recording error, which will be discussed in Section 6.6. With recording error, an
incorrectly recorded vehicle at one observation station is independent of it being
incorrectly recorded at another observation station.
The percentage of vehicles that are unidentifiable due to missing, damaged,
dirty, covered, or blocked license plates may vary from roadway to roadway; however, it
is not expected to be large. The estimate made as a result of this study from several
roadway observations is approximately 5% of all vehicles.

6.3.1 Time-Length of License Plate Legibility

The length of time a license plate on a passing vehicle is legible to an observer
standing on the side of the road depends upon: 1) the speed of the vehicle, 2) the visual
acuity of the observer, and 3) the size and style of the digits on the license plate. Visual
acuity is stated as 20/X, where 20 is the distance (in feet) at which you can see an
object, and X is the distance a person with “normal” vision can see that same object.
Therefore, a person with 20/20 vision has normal vision. A denominator less than 20
means better-than-average visual acuity, and a denominator greater than 20 means
worse-than-average visual acuity. At 20/200, a person is considered legally blind.
20/20 visual acuity is defined as the distance at which a person can see an
object that spans 5 minutes of arc (5’) or 1/12 of a degree. Figure 38 illustrates this.
 88

Figure 38: Definition of 20/20 Visual Acuity

Therefore, by solving for d in Figure 38, a person with 20/20 visual acuity will
theoretically be able to identify the characters on a license plate (approximately 2.75” in
height) at a distance of 158 feet. A person with 20/10 vision will be able to see the
characters at twice that distance (316 feet), and a person with 20/40 vision will be able to
see the characters at half that distance (79 feet).
Depending upon the speed of the vehicle and the visual acuity of the observer, a
license plate will remain legible for a short period of time after it passes the observer.
Table 8 summarizes these times.

Table 8: Theoretical Time-Length of License Plate Legibility (2.75” Characters)

Visual Distance Vehicle Speed (mph)
Acuity Leg. (ft) 5 15 25 35 45 55 65 75
20/10 316 43.1 14.4 8.6 6.2 4.8 3.9 3.3 2.9
20/15 211 28.7 9.6 5.7 4.1 3.2 2.6 2.2 1.9
20/20 158 21.5 7.2 4.3 3.1 2.4 2.0 1.7 1.4
20/30 105 14.4 4.8 2.9 2.1 1.6 1.3 1.1 1.0
20/40 79 10.8 3.6 2.2 1.5 1.2 1.0 0.8 0.7

It can be seen, then, that a license plate string is legible to an observer with
20/10 vision on a vehicle traveling at 5 mph for 43.1 s after the vehicle passes.
However, that same license plate string is only legible to an observer with 20/40 vision
for 10.8 s. Likewise, a license plate string is legible to an observer with 20/20 vision on a
 89

vehicle traveling at 5 mph for 21.5 s. However, a license plate is only legible to the
same observer (with 20/20 vision) for 1.4 s if the vehicle is traveling 75 mph. This table
illustrates that time of legibility decreases as vehicle speed increases and visual acuity
decreases.
Many of the Indiana license plates also use characters that are half the size as
the normal characters. In many cases, these characters also have to be identified and
recorded in accordance with the rules described in Section 6.1. Therefore, Table 9
illustrates the legibility distance for these half-sized characters and the length of time
they are legible from the roadside.

Table 9: Theoretical Time-Length of License Plate Legibility (1.25” Characters)

Visual Distance Vehicle Speed (mph)
Acuity Leg. (ft) 5 15 25 35 45 55 65 75
20/10 143 19.5 6.5 3.9 2.8 2.2 1.8 1.5 1.3
20/15 95 13.0 4.3 2.6 1.9 1.4 1.2 1.0 0.9
20/20 72 9.8 3.3 2.0 1.4 1.1 0.9 0.8 0.7
20/30 48 6.5 2.2 1.3 0.9 0.7 0.6 0.5 0.4
20/40 36 4.9 1.6 1.0 0.7 0.5 0.4 0.4 0.3

After calculating the table above and performing some field data collection, it was
realized that the characters on license plates that were the most difficult to read (the
standard license plate shown in Figure 32) were, in fact, not legible at these distances.
Therefore, knowing that the visual acuity of the researcher is 20/15, the legibility
distance of standard Indiana license plates were measured on stationary vehicles. The
legibility distance of the small characters on the standard Indiana passenger vehicle
license plate was 72 feet (instead of 95 feet calculated in the table above). Therefore,
the legibility distances for other visual acuities were determined and the time lengths of
license plate legibility were recalculated using these values.

Table 10: Observed Time-Length of License Plate Legibility (1.25” Characters)

Visual Distance Vehicle Speed (mph)
Acuity Leg. (ft) 5 15 25 35 45 55 65 75
20/10 108 14.7 4.9 2.9 2.1 1.6 1.3 1.1 1.0
20/15 72 9.8 3.3 2.0 1.4 1.1 0.9 0.8 0.7
20/20 54 7.4 2.5 1.5 1.1 0.8 0.7 0.6 0.5
20/30 36 4.9 1.6 1.0 0.7 0.5 0.4 0.4 0.3
20/40 27 3.7 1.2 0.7 0.5 0.4 0.3 0.3 0.2
 90

There are several reasons the characters are not legible at the theoretical
distances calculated in Table 9. First, some characters are easier to see than others.
Also, the characters do not appear on a single solid background color. More importantly,
the characters are thinner (skinnier) than characters used to measure visual acuity.
Finally, the visual acuity of the observer may not be exactly 20/15.

6.3.2 License Plate Identification Time

After the license plate identification times were measured in the field, the data
was analyzed to determine if the speed of the vehicle had any effect on the license plate
identification time. Table 11 illustrates the means and standard deviations of the
identification time for each parameter.

Table 11: Identification Time (s) by Vehicle Speed

4-Character Full String
Speed Mean St. Dev. Mean St. Dev.
15 mph 0.77 0.16 1.36 0.28
35 mph 0.75 0.13 1.02 0.21
55 mph 0.86 0.16 1.08 0.23

It can be seen from the table that, in general, the mean identification time
increases as vehicle speed increases. The primary exception to this was the mean
identification value of a full string at 15 mph (1.36 s), although the 4-character
identification time for the same parameters also seemed slightly high at 0.77 s. These
higher values may be explained by the fact that these values were the first to be
measured in the field, and the researcher became slightly better at identifying vehicles
over time. Another possibility is that, at lower vehicle speeds, the time length of legibility
is longer, so identifying vehicles quickly is not as critical as it is at higher vehicle speeds
because the observer has enough time to take a second look at the license plate.

6.3.3 Probability of String Identification

Because the ID times are approximately normally distributed, the probability that
a passing license plate will be recorded based on vehicle speed and visual acuity can be
determined by plotting the time lengths of legibility calculated in Table 10 on the ID time
 91

distributions using the means and standard deviations in Table 11 and obtaining the
percentiles. These percentiles are the probability that a license plate will be recorded at
a given speed.
However, because it is known that the mean identification time varies with
vehicle speed, different distributions must be drawn for different vehicle speeds. The
mean identification time for each vehicle speed was determined by conducting a linear
regression on the mean identification times collected in the field. (The full string mean
value at 15 mph was not included in the regression because it was considered unusually
high.)
Figures 39 through 40 below illustrate the probability of identifying a standard
issue Indiana passenger vehicle license plate (with 1.25” characters) on a vehicle at any
speed for five different visual acuities ranging from 20/10 to 20/40. Figure 39 is for 4-
character string identification and Figure 40 is for full string identification.

Probability of 4-Character Identification

24' Initial Viewing Distance

1.0
0.9
0.8
0.7
20/10
Probability

0.6 20/15
0.5 20/20
0.4 20/30
20/40
0.3
0.2
0.1
0.0
0 10 20 30 40 50 60 70
Vehicle Speed (mph)

Figure 39: Probability of 4-Character Identification

 92

Probability of Full String & State Identification

1.0

0.9

0.8
Probability of Identification
0.7
20/10
0.6
20/15
0.5 20/20
0.4 20/30
20/40
0.3

0.2

0.1

0.0
0 10 20 30 40 50 60 70
Vehicle Speed (mph)

Figure 40: Probability of Full String Identification

These figures were constructed based on the analysis of mean identification

times for one researcher. Naturally, variations in identification time also exist from
person to person. Initially, it was determined that data on identification times from other
persons be obtained. However, because of the complexity involved in measuring
identification times using a self-operated stopwatch, the difficulty in recruiting a variety of
persons for the task, and the length of time required to record the data, data was only
obtained from one person.

6.3.4 Maximum Vehicle Speed for Manual License Plate Data Collection
In the previous section, the identification time, which is the time required to locate
and memorize a license plate string, was determined. This identification time should not
exceed the time-length of legibility for any given roadway speed and visual acuity. If so,
the license plate string on a vehicle traveling away from the observer will become
illegible before the observer has enough time to identify the string. Therefore, we can
calculate the maximum vehicle speed at which an observer with a specific visual acuity
should attempt to record license plate strings using manual methods.
Table 12 recommends the vehicle speed above which manual methods
(clipboard, audio, and laptop) should not be used. Instead, the video method should be
 93

used. These speeds in Table 12 are the points at which the probability curve for each
visual acuity begins to drop from 1.0.

Table 12: Vehicle Speed (mph) above which Video Method is Recommended
1.25" Characters 2.75" Characters
Visual Acuity 4-Character Full String 4-Character Full String
20/10 45 35 100 70
20/15 30 20 65 45
20/20 20 10 45 30
20/30 5 n/a 25 20
20/40 n/a n/a 15 10

It can be seen in Table 12 that, even with 20/10 visual acuity, recording 4-
character license plates with a manual method is not recommended for vehicle speeds
greater than 45 mph. For an observer with 20/20 visual acuity, the maximum speed is
20 mph. For full string recording, manual methods may be used for vehicle speeds less
than 10 mph and visual acuity of 20/20, or less than 30 mph for visual acuity of 20/10.
Again, this is only recommended for recording the near lane.
For vehicle speeds that exceed the maximum speed for identification with a
manual method, video recording is likely the best method. Video recording was
discussed in greater detail in Chapter 6.
If conducting an OD study on a subset of vehicles, such as trucks, the video
method is recommended. This is because trucks may have multiple license plates
mounted in different locations, and are often dirty and harder to see than passenger
vehicle license plates. These factors generally make the identification time of a license
plate on a truck higher than that of a passenger vehicle.
If video recording is not possible, other alternatives exist. For example, two
personnel could be used at each station – one for observing vehicles with binoculars,
and the second for recording the license plate strings. By doing this, the length of time a
license plate is legible increases significantly, but twice as many people are needed in
the field. This was not evaluated as part of this project. In addition, it may be possible to
reduce the vehicle speed on the road through reduced speed limits (using work zone
traffic control methods), but this would require permission and assistance from the DOT.
 94

In addition, police enforcement may be needed to ensure that motorists actually reduce
speed enough to make the manual method successful.

6.4 License Plate Recording

From the above paragraphs, the maximum vehicle speed at which license plates
can be identified was found for any visual acuity. Just as the identification time was
determined in a preceding section, the recording time is also desired, which is the
amount of time the observer takes to record the plate just identified into some recording
medium. The recording time will differ for each of the recording methods (clipboard,
audio, and laptop). Because the average recording time for one method is likely to be
different than that of the other methods, each of the methods is likely to be limited by a
minimum vehicle headway (and, conversely, maximum vehicular flow rate) at which a
second vehicle will pass the observation point before the license plate of the previous
vehicle is recorded, thereby missing the second vehicle. Furthermore, the recording
time for a 4-character string is likely less than that of recording a full license plate string,
so depending on the technique used (matching or follow-up survey), one method may
not be suitable for both techniques. The objective of determining these recording times
is to identify the method that enables the highest capture rate to be obtained. The
capture rate is the number of license plates recorded divided by the total number of
vehicles passing the observation station.

6.4.1 License Plate Recording Time

In order to determine these recording times, a procedure was used that was
similar to that for measuring string identification times discussed earlier. In this case, the
identification and recording times were measured together.
This time was measured by a researcher using a self-operated stopwatch. The
timer began when the license plate entered the researcher’s field-of-view and remained
running until the last digit of the license plate string was recorded. This procedure was
repeated 50 times, recording 4-character and full license plate strings by each method.
Table 13 illustrates the recording time mean and standard deviation for each
method.
 95

Table 13: Recording Time (s) by Method

4-Character Full String
Method Mean St. Dev. Mean St. Dev.
Clipboard 2.99 0.33 4.95 0.81
Audio 1.76 0.25 2.84 0.50
Laptop 2.51 0.49 5.38 1.05

The average time to identify and record a 4-character license plate to a clipboard
was 2.99 s. Therefore, if the license plate of vehicle 1 is being identified and recorded,
and vehicle 2 is following vehicle 1 with headway of less than the recording time, the
license plate of vehicle 2 will not be recorded because the observer has not completed
recording of vehicle 1.

6.4.2 Vehicle Interarrival Times (Headways)

The probability that a vehicle will have a headway greater than a given value (the
recording time) can be calculated by the equation P(T ≥ t) = e-λt, where λ is the flow rate
in vehicles per second and t is the headway in seconds. This equation is derived from
the Poisson model of traffic flow and has a negative exponential distribution (Fricker &
Whitford, 2004).
For example, if the flow rate of a particular road is 400 veh/h (0.111 veh/s) and
the time to record a license plate is 2.99 s, the probability that the headway between
vehicles is greater than 2.99 s is 71.8%.

6.4.3 Maximum Flow Rate for Manual Methods

The objective of this task is to specify a flow rate at which the video method
should be used instead of manual methods (clipboard, audio, and laptop), because at
higher flow rates, a higher percentage of vehicles will be missed. To do this, the joint
probability density function was approximated. This was done by randomly selecting a
recording time (assumed to be normally distributed and based on the mean and
standard deviations in Table 13) and a vehicle headway (based on the negative
exponential distribution in Section 7.4.2). These two values were compared, and the
process was repeated 10,000 times for various flow rates. The percentage of the time
that the recording time was less than the headway was determined. Figure 41 illustrates
the percentage of vehicles that could be recorded, based on flow rate for each method.
 96

Percent of Vehicles Captured

1.00

0.90
0.80

0.70 Clip-4

0.60 Clip-Full
Probability

Aud-4
0.50
Aud-Full
0.40
Lap-4
0.30 Lap-Full
0.20
0.10
0.00
0

200

400

600

800

1000

1200

1400

1600

1800

2000
Flow Rate (veh/h)

Figure 41: Capture Rates by Vehicle Flow for each Method

Using Figure 41, the percent of vehicles captured can be determined for any flow
rate. If the capture rate is sufficient for the purpose of the study, that method can be
used. If not, however, the video method is recommended. Table 14 summarizes the
maximum flow rates for the 95%, 85%, and 75% capture rates (rounded to the nearest
25 vehicles).

Table 14: Maximum Flow Rate (vphpl) using Manual Recording Methods
95% Capture 85% Capture 75% Capture
Method 4-Char Full 4-Char Full 4-Char Full
Clipboard 50 25 200 125 350 225
Audio 100 50 350 225 650 375
Laptop 75 25 250 100 450 200

Table 14 illustrates that, in order to record 4 characters on 95% of all identifiable

license plates using the audio method, the flow rate cannot exceed 100 vphpl. If the
acceptable capture rate is reduced, the maximum flow rates increase significantly. For
flow rates that exceed those shown in the table, the video method is recommended for
recording license plates.
 97

This table only specifies flow rates for uninterrupted flow. If license plate data is
being collected from a roadway with interrupted flow (e.g., downstream from a traffic
signal), vehicles are more likely to be spaced closely together. This depends on the
amount of queueing at the signal. If there is a lot of queueing and vehicles are traveling
past the observation station in platoons, the video method is recommended for data
collection. In addition, this table does not account for observer fatigue.
Chapter 8 evaluates how the capture rates affect the overall accuracy of the
estimated OD pairs.

6.5 Video Method

Based on the preceding sections, manual recording of license plate strings with
the clipboard, audio, and laptop methods may not suffice in situations with high vehicle
speeds and/or high flow rates. In these cases, video can be used to record the license
plates. The video can then be played back at a controlled speed in the lab to obtain
more license plate strings than would be obtained using any other method for that
roadway. However, the legibility of license plate strings is dependent on the quality of
the video. There are a number of issues to consider when using a camcorder to obtain
the best possible video quality, including the distance from the roadside, the angle, and
zoom. In addition, lighting conditions, vehicle speeds, and flow rates should also be
considered. These issues were discussed in detail in Chapter 6. The goal of the
following section is to evaluate the process of extracting the appropriate information from
the clipboard, audiotape, laptop, videotape after field collection has been completed.

6.6 Errors in License Plate Data

Obviously, the objective in recording license plates is to record as many as
possible accurately. However, it is unlikely that an observer will make no mistakes in
recording the data in the field. While the capture rate was discussed in the previous
section, the error in recording these license plates is discussed in this section.

6.6.1 Field and Transcription Errors

Several experiments were conducted to determine the field and transcription
errors for each of the data collection methods. Field errors are errors that are
encountered in the field as the license plate strings are identified and recorded from the
 98

vehicle onto the clipboard, audio tape, laptop, or videotape. Transcription errors, on the
other hand, are errors that are made after the data collection period while transferring
the data in its raw form on the recording medium into some database.
For each of the clipboard, audio, and laptop methods, 100 4-character and full
license plate strings were recorded. During each of these sessions, a camcorder was
also set up and taped the same set of vehicles on the roadway. At the end of each
session, the data recorded on the clipboard, audio tape, and laptop were transcribed into
separate spreadsheets. Likewise, the video was reviewed, and all license plates strings
were transcribed into the appropriate spreadsheets. This transcription process was
repeated for each session. In the end, the spreadsheet for each session contained each
license plate four times: two columns of the license plate records from the clipboard or
audio tape, and two columns from the analysis of the corresponding video. In the case
of the laptop, there are only three columns because the license plates are entered
directly into the computer in the field and do not require transcription.
By transcribing all of the data twice, the field and transcription errors, as well as
transcription time can be determined for each of the methods. These errors can occur
when the license plate is misread. An example of this type of error is mistaking an “O”
for a “Q”, “6” for an “8”, and “M” for an “N”. These errors can also occur when the
characters are identified correctly, but then incorrectly recorded. Examples of this type
of error include transposing letters (“68” instead of “86), speaking too quickly in the audio
recorder, and incorrect key entry on the laptop. Table 15 summarizes the errors for
each of the methods.

Table 15: Field & Transcription Errors by Method

Field Transcription
Method 4-Char Full 4-Char Full
Clipboard 0.0% 3.0% 0.0% 1.5%
Audio 3.0% 4.0% 1.5% 0.5%
Laptop 2.9% 3.7% n/a n/a
Video n/a n/a 0.3% 0.7%

From the table above, the total amount of error for any particular method after
field recording and transcribing the data is the sum of the two values shown. For
example, if full strings of the license plates are field recorded using the audio method
 99

and subsequently transcribed, the total amount of error will be 4.5% (4.0% from the field
and 0.5% from the transcription). These errors, in effect, reduce the capture rates even
further, and must be considered when expanding the sample data to the entire
population. This is discussed in Chapter 8.
The 4-character recording error for all methods is generally less than the full
string recording error. It seems logical that an observer is more likely to make a mistake
when recording more characters. To illustrate, it is generally easier to remember the last
four digits of a phone number than it is the entire phone number.
The clipboard method had the least amount of field error among the three
methods. Generally, fewer errors are made with the clipboard than when keying entries
into a laptop. However, unlike the clipboard and audio methods, the laptop does not
have an extra transcription step, which adds another element of error to the first two
methods.
The audio method had the most field error of the three methods. This could be
because, unlike the clipboard or laptop methods, the audio method does not require the
observer to take his or her eyes off the traffic, thus enabling him or her to continuously
record license plates. In some cases, the license plates appear faster than the observer
should attempt to record, which may cause the observer to make more of the mistakes
described above, or make transcription difficult because the characters are recorded too
quickly to understand during playback.
For the laptop method, during field recording, license plate entries were keyed
into the laptop using the standard laptop keyboard with the numbers along the top.
During the transcription process, license plate entries were keyed in using a keyboard
with a number pad. This partially explains why the errors for field recording with a laptop
are greater than the transcription errors for all other methods. Generally, it is easier to
type numbers on a number pad rather than along the top of the keyboard, and
transcription is done indoors in a more comfortable environment.
These field recording errors were measured under low speed conditions, and do
not reflect errors that may be encountered if manual recording is conducted on a speed
higher than those recommended by Table 12. In such cases, the error will likely be
significantly higher, which reduces the overall accuracy when expanding the data.
Transcription errors for each of the methods are generally lower than field errors.
These errors are generally caused by poor handwriting (for the clipboard method), poor
 100

audio such as background noise (for the audio method), and poor video quality (for the
video method). Again, license plates recorded by a laptop in the field do not require
transcription.
When these errors occur, there is the possibility (for the license plate matching
technique) that some license plates that should be matched to another location will not
be matched. Therefore, the number of matches need to be increased to reflect these
errors. Likewise, for the follow-up survey technique, the license plate will not provide the
correct vehicle owner address (or an address at all) by searching the BMV database.
Therefore, the number of surveys sent out will be less than the number of license plates
recorded in the field, which will reduce the overall accuracy of the survey results.
While the true error is not known in an actual OD study, the values summarized
in Table 15 may be used as a guideline when expanding the data only if, for manual
methods, the data was not collected from vehicle speeds that exceeded those
recommended in Table 12.

6.6.2 Transcription Time

In addition to the variation in errors among the methods, the amount of time
required to transcribe the license plates varies significantly from method to method. This
does not influence the overall quality of the results, but certain methods will require more
time (and more cost). Automatic transcription can be used to reduce this transcription
time. However, the added expense of automatic transcription technology and the
generally lower capture rate must also be considered. Again, automatic transcription
technologies were not evaluated as part of this project.
The total transcription time for each of the methods was simultaneously
measured from the same license plate data used to determine errors. Table 16
summarizes these transcription times. The coefficients that appear in Table 16 were
determined by dividing the transcription time by the number of license plates transcribed.
The table does not account for long-term effects, such as fatigue, but are recommended
for estimating transcription times.
 101

Table 16: Transcription Time (sec) by Method

Method 4-Char Full
Clipboard 2.4n 4.2n
Audio L + 0.3n L + 1.2n
Laptop n/a n/a
Video L + 3.5n L + 6.3n

For the clipboard method, the transcription rate is simply the total time required to
transcribe n license plates divided by the number of license plates. Therefore, if 10,000
4-character license plates are recorded during an OD study, it will take approximately
6.7 hours (2.4*10,000/3600) to manually transcribe them.
For the audio and video methods, the transcription time required consists of L,
which is the total time length (in seconds) of audio or video tape recorded in the field
plus the average time paused per license plate times the number of license plates
recorded. For example, if 10,000 4-character license plates were recorded on 8 hours of
videotape, the approximate manual transcription time is 17.7 hours
([8*3600+3.5*10,000]/3600).
Generally, the video method has the highest transcription time. However, it must
be noted that video also has higher capture rates and lower errors than the other
methods.
 102

CHAPTER 7 – ACCURACY OF OD STUDY TECHNIQUES

While the previous chapter discusses traffic conditions as they relate to a license
plate survey, it is often unknown how accurate the results are once the data is analyzed.
This chapter describes methods to determine sample size for vehicle intercept and
license plate follow-up surveys. In addition, the license plate matching technique was
simulated at varying capture rates and the vehicle tracing technique was simulated at
varying probe samples to determine the effect on the overall accuracy of the estimation.
The simulation was conducted using Microsoft Excel. The details of the simulation for
each technique will be discussed in greater detail below.

7.1 License Plate Matching Technique

In the literature review, only one source (Schaefer, 1988) was found that
explicitly defined the number of license plates that should be recorded at a roadside
station for the license plate matching technique for the purpose of estimating the
proportion of traffic at one station that has a particular destination. The equation is
simply the required sample size n to estimate a proportion p, given a particular level of
confidence z and error e. This equation is n = pqz²/e², where q = 1 – p.
If an estimate of the proportion p cannot be determined, the most conservative
approach is to set p = 0.5. The sample size n for 95% confidence (z = 1.96, two-tailed)
and 5% error is n = (0.5)*(1-0.5)*(1.96²)/(0.05²) = 384 license plate observations.
While the equation is simple, it does not consider the volume of traffic from which
the license plates are being sampled. As the traffic volume increases, the capture rate
usually decreases, which reduces the number of matches obtained in the sample. Then,
in expanding the sample to the population, the estimate for the proportion (and
corresponding number of vehicles) can have high variability and error.
 103

For example, consider the situation in which the actual traffic volume at Station 1
and Station 2 is 10,000 vehicles, and the true proportion p equals 0.25, which means the
total number of vehicles that pass both Stations 1 and 2 is 2,500 (0.25*10,000). By
capturing 384 vehicles at each station (3.8% capture rate), a number of matches will be
obtained. This situation was simulated 30 times, in which the number of matches
obtained in the sample ranged from 0 to 9. These samples were then expanded to
estimate the actual proportion p and number of vehicle passing both stations. While the
average p of these estimates was 25%, the standard deviation was 14.5% (1,454
vehicles). The estimates ranged from 0% (0 vehicles) to 61% (6,104 vehicles), which is
a range of -2,500 to 3,604 vehicles.
This example illustrates the variability that can be encountered in the license
plate matching technique and the importance of collecting as many license plates as
possible.
Even under the best circumstances, however, it is unlikely that all license plates
on a particular road can or will be recorded during an OD study. In the matching
technique, the license plate must be recorded correctly two separate times at two
separate locations for that OD pair to be recognized. Because of this, missing license
plates has the potential to introduce a significant amount of error. Therefore, it was
desired to simulate various scenarios to determine how the quantity of the data recorded
influences the accuracy of the projected results.

7.1.1 Simulation
As a starting point for this analysis, assume there are two recording stations
(Station 1 and Station 2). A total of 1000 vehicles, each with a unique identifier (license
plate), pass by each station. Assume that 25% of all vehicles (250 vehicles) that pass
Station 1 also pass Station 2. The true proportion p between stations is 25%. This
percentage and number of vehicles are the values sought during an OD study. If 100%
of vehicles passing by both stations are captured, and all of the license plates at both
stations are recorded correctly, 250 matches would be found between the stations and
there would be zero error. However, this is an ideal situation and highly unlikely.
Knowing that the capture rates are never 100%, the capture rates were varied at
each station. Vehicles were randomly selected from each station to simulate the capture
rate. The two stations were then compared and the number of matches between the
 104

stations was noted. Because it is assumed that there were no recording errors, the
number of matches was divided by the capture rates at both stations to estimate the total
number of trips between the two stations during the study period.
For example, assume again that a total of 1,000 vehicles pass both Station 1 and
Station 2, and that 25% of vehicles passing Station 1 also pass Station 2. Now assume
that 75% of vehicles passing Station 1 and 50% of vehicles passing Station 2 have their
license plates captured in the field. By comparing the capture lists from each station
(which were randomly selected from the full lists of all vehicles passing each station), the
number of matches is found, then divided by 0.375 (75%*50%) to estimate the
proportion and total number of trips between the two stations. The estimated number of
trips is compared to the actual number of trips (in this case, 25% of 1000 vehicles, or
250 vehicles), and the percent error between the estimated and actual trips is calculated.
The process of randomly selecting the percent of vehicles captured from each station,
estimating the total number of trips, and calculating the percent error was conducted 50
times for each true proportion and capture rates at both stations. The standard deviation
of the percent error was then calculated for each set. For this particular combination, the
standard deviation after 50 simulations was 7.2%. This means that, with 95%
confidence, the estimated number of trips for a true proportion of 25% is within ±14.1%
of the true number of trips.

7.1.2 Varying Capture Rates, Proportions, and Traffic Volumes

The simulations described above only account for one combination of station
capture rates, proportions, and station traffic volumes. Therefore, these variables were
changed at certain intervals to evaluate a wide range of possible combinations.
First, using the same traffic volumes of 1,000 vehicles at each station, a true
proportion of 1% between stations, the capture rates at each station were varied from
50%, 75%, and 95%. The capture rates at each station were held equal for simplicity,
although in reality, they are independent of one another. Because expanding the data is
based on the product of the capture rates, however, relative capture rates have little or
no effect on the estimation error. In the end, 50 simulations were run on each of the 3
combinations of capture rates.
 105

After these simulations, the true proportion p between the stations was changed
to 5%. Then, 50 simulations of the 3 combinations of capture rates were repeated. This
was also repeated for a true proportion of 25% and 50%.
Finally, the traffic volumes at both stations were changed, during which 50
simulations were run on all combinations of true proportions and capture rates. The
traffic volumes at the stations were changed to 5,000 and 10,000 vehicles.

7.1.3 Results
Figures 42 – 45 below illustrate the change in error as the true proportion of
traffic between stations changes. Figure 42 illustrates the error for a proportion of 1%,
and Figures 43 – 45 illustrate the error for proportions of 5%, 25%, and 50%,
respectively. On each figure, the product of capture rates is shown on the x-axis, and
each line represents a different traffic volume at the observation stations (assuming the
traffic volume is the same at both stations for simplicity).

Error in Trip Estimation

for Station Traffic Volume of 100 Vehicles

100%
25.0%
90%
56.3%
Standard Deviation of Percent Error

80% 90.3%
(after 50 simulations)

70%

60%

50%

40%

30%

20%

10%

0%
0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50%
Proportion

Figure 42: Error by Proportion for Traffic Volume of 100 Vehicles

 106

Error in Trip Estimation

for Station Traffic Volume of 1,000 Vehicles

60%
25.0%
Standard Deviation of Percent Error 56.3%
50%
90.3%
(after 50 simulations)

40%

30%

20%

10%

0%
0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50%
Proportion

Figure 43: Error by Proportion for Traffic Volume of 1,000 Vehicles

Error in Trip Estimation

for Station Traffic Volume of 5,000 Vehicles

30.0%
25.0%
56.3%
Standard Deviation of Percent Error

25.0%
90.3%
(after 50 simulations)

20.0%

15.0%

10.0%

5.0%

0.0%
0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50%
Proportion

Figure 44: Error by Proportion for Traffic Volume of 5,000 Vehicles

 107

Error in Trip Estimation

for Station Traffic Volume of 10,000 Vehicles

14% 25.0%
Standard Deviation of Percent Error 56.3%
(after 50 simulations) 12% 90.3%

10%

8%

6%

4%

2%

0%
0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50%
Proportion

Figure 45: Error by Proportion for Traffic Volume of 10,000 Vehicles

Figures 42 – 45 above illustrate that, as the proportion between stations

increases, the error in the estimated number of trips decreases. In any real situation, the
proportion between any OD pair is seldom greater than 50%. The points on the figures
represent the proportions that were simulated. A product of capture rates of 25% is a
50% capture rate at each station. Likewise, 56.3% is the product of 75% capture at
each station, and 90.3% is the product of 95% capture at each station. As stated earlier,
the capture rates were held equal at each station, however, a product of 56.3% could
also be obtained by station capture rates of 100% and 56.3%, 85% and 66%, etc.
To describe the figures, assume, for example, 75% of license plates are recorded
at the stations with traffic volumes of 5,000 vehicles, and 145 matches are found
between the stations. The estimated total number of vehicles traveling between these
two stations is determined by dividing the sample by the product of the capture rate.
The estimated number of trips between the stations is 145/0.563 = 258 vehicles. The
estimated proportion of all traffic at Station 1 traveling to Station 2 is 258/5,000 = 0.051.
The standard deviation for a proportion of 5.1% is approximately 5.5% (from Figure 44).
Therefore, the 95% confidence interval is ±10.8% error in the true number of trips
between the stations (1.96*5.5%). Therefore, the true number of trips is between 230
and 286 (258 ± 258*.108), and the true proportion is between 4.6% and 5.7%.
 108

Figures 46 – 49 illustrate the same information presented in Figures 42 – 45

above. However, the x-axis is traffic volume and each line represents a different capture
rate product. Figure 46 illustrates errors for a proportion of 1%, while Figures 47 – 49
present errors for proportions of 5%, 25%, and 50%, respectively.

Error in Trip Estimation

for a True Proportion of 1%

100%

90%
Standard Deviation of Percent Error

80%
(after 50 simulations)

70%

60%
25.0%
50% 56.3%
90.3%
40%

30%

20%

10%

0%
0 2000 4000 6000 8000 10000
Traffic Volume at Observation Stations

Figure 46: Error by Traffic Volume for a True Proportion of 1%

Error in Trip Estimation

for a True Proportion of 5%

70%

60%
Standard Deviation of Percent Error

50%
(after 50 simulations)

40% 25.0%
56.3%
30% 90.3%

20%

10%

0%
0 2000 4000 6000 8000 10000
Traffic Volume at Observation Stations

Figure 47: Error by Traffic Volume for a True Proportion of 5%

 109

Error in Trip Estimation

for a True Proportion of 25%

35%

30%
Standard Deviation of Percent Error

25%
(after 50 simulations)

20% 25.0%
56.3%
15% 90.3%

10%

5%

0%
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
Traffic Volume at Observation Stations

Figure 48: Error by Traffic Volume for a True Proportion of 25%

Error in Trip Estimation

for a True Proportion of 50%

20%

18%
Standard Deviation of Percent Error

16%

14%
(after 50 simulations)

12%
25.0%
10% 56.3%
90.3%
8%

6%

4%

2%

0%
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
Traffic Volume at Observation Stations

Figure 49: Error by Traffic Volume for a True Proportion of 50%

Figure 46 – 49 show that, for any given product of capture rates, the error
decreases (at a decreasing rate) as traffic volumes increase. For example, Figure 49
shows that a 90.3% capture rate product has approximately the same accuracy when
 110

sampled on a roadway with 5,000 vehicles and 10,000 vehicles. The absolute number
of vehicles sampled, however, is much higher for the higher volume road.
Figures 50 – 53 illustrate the percent error in the estimated trips for station traffic
volumes of 100, 1,000, 5,000, and 10,000 vehicles, respectively.

Error in Trip Estimation

for Station Traffic Volume of 100 Vehicles

100%

90% 1% Proportion
5% Proportion
Standard Deviation of Percent Error

80%
25% Proportion
70% 50% Proportion
(after 50 simulations)

60%

50%

40%

30%

20%

10%

0%
20% 30% 40% 50% 60% 70% 80% 90% 100%
Product of Capture Rates

Figure 50: Error by Capture Rate Product for Traffic Volume of 100 Vehicles

Error in Trip Estimation

for Station Traffic Volume of 1,000 Vehicles

60%
1% Proportion
5% Proportion
Standard Deviation of Percent Error

50%
25% Proportion
50% Proportion
(after 50 simulations)

40%

30%

20%

10%

0%
20% 30% 40% 50% 60% 70% 80% 90% 100%
Product of Capture Rates

Figure 51: Error by Capture Rate Product for Traffic Volume of 1,000 Vehicles
 111

Error in Trip Estimation

for Station Traffic Volume of 5,000 Vehicles

30%
1% Proportion

Standard Deviation of Percent Error

5% Proportion
25%
25% Proportion
50% Proportion
(after 50 simulations)

20%

15%

10%

5%

0%
20% 30% 40% 50% 60% 70% 80% 90% 100%
Product of Capture Rates

Figure 52: Error by Capture Rate Product for Traffic Volume of 5,000 Vehicles

Error in Trip Estimation

for Station Traffic Volume of 10,000 Vehicles

14% 1% Proportion
Standard Deviation of Percent Error

5% Proportion
12% 25% Proportion
(after 50 simulations)

50% Proportion
10%

8%

6%

4%

2%

0%
20% 30% 40% 50% 60% 70% 80% 90% 100%
Product of Capture Rates

Figure 53: Error by Capture Rate Product for Traffic Volume of 10,000 Vehicles

It can be seen in Figure 50 that, for any given proportion, the percent error
decreases as the product of capture rates increase. This is because, on average, the
sample will contain less variability in the number of matches obtained, and the number of
matches is not expanded as much. Table 17 summarizes the data used to build Figures
42 – 53.
 112

Table 17: Standard Deviation of Percent Error in Trip Estimations

1% Proportion 5% Proportion
CR Traffic Volume CR Traffic Volume
Prod. 100 1000 5000 10000 Prod. 100 1000 5000 10000
25.0% 177.2% 58.1% 25.8% 13.4% 25.0% 70.9% 19.4% 11.2% 7.3%
56.3% 85.1% 28.0% 11.8% 10.0% 56.3% 35.2% 12.4% 5.5% 3.8%
90.3% 21.9% 9.4% 4.2% 2.5% 90.3% 12.8% 4.8% 1.8% 1.5%
100% 0% 0% 0% 0% 100% 0% 0% 0% 0%
25% Proportion 50% Proportion
CR Traffic Volume CR Traffic Volume
Prod. 100 1000 5000 10000 Prod. 100 1000 5000 10000
25.0% 31.4% 9.7% 4.6% 3.5% 25.0% 19.2% 6.5% 2.8% 2.1%
56.3% 18.1% 4.8% 1.7% 1.9% 56.3% 9.3% 2.7% 1.0% 0.9%
90.3% 5.5% 1.3% 0.7% 0.6% 90.3% 3.4% 1.1% 0.5% 0.3%
100% 0% 0% 0% 0% 100% 0% 0% 0% 0%

The capture rates become a bigger issue when the OD volumes of a subset of
vehicles, such as trucks, are being determined, because the traffic volume (number of
trucks observed) during the study period may be low, which increases the error in the
estimation. In this case, it is important that high capture rates are achieved. Because of
the factors cited in Chapter 7 and the lower accuracy mentioned above, the video
method is recommended for recording license plates in truck OD studies.

7.2 Vehicle Intercept and License Plate Follow-Up Survey Techniques

The accuracy of vehicle intercept surveys and license plate follow-up surveys for
a given station depends on the number of completed surveys at a particular station. For
vehicle intercept surveys, a higher percentage of surveys (generally >90%) will be
completed in the roadside interview method than with the postcard questionnaire
method. In order to obtain the same amount of information as the roadside interview
method, the postcard questionnaires will have to be distributed to more vehicles. In
terms of response rates, postcard questionnaires are much more similar to a license
plate follow-up survey, because the surveys are completed at the will of the driver,
generally 15% – 30%, depending on the length and format of the questionnaire (Virkud,
1995). For license plate follow-up surveys, full license plate numbers have to be
 113

captured in the field and matched with an address at the BMV, which increases the
number of license plates that must be recorded.
While the Travel Survey Manual (TMIP, 1996) provides sample size equations for
household travel surveys, it does not explicitly provide any details for determining
sample sizes for vehicle intercept surveys. In searching the literature for sample size
equations, two sources were found, as discussed below.
The first equation was found in a document entitled “Traffic Surveys by Roadside
Interview” (The Highways Agency, 1992) published by several agencies in the United
Kingdom. The equation is:

pqN 3
n=
⎡⎛ E ⎞ 2 ⎤
⎢⎜ ⎟ ( N − 1) + pqN ⎥
2

⎢⎣⎝ Z ⎠ ⎥⎦

n = number of completed surveys required

p = estimated proportion of total traffic at the station with a particular destination
q=1–p
E = accuracy or absolute error (percent error * N)
N = traffic volume at survey station
Z = normal variate for specified level of confidence (eg. 1.96 for 95% confidence)
Stokes et al. (1989) use another equation proposed by Hajek (1977) that
provides the sampling rate required to develop external-external trip tables based on the
estimated proportion of traffic for a particular OD pair, the desired accuracy and
confidence level, and the volume of traffic at the station. The equation is shown below:

Z 2 pq
r=
[ (
(N − 1)W 2 + Z 2 pq )]

r = sample rate
p = estimated proportion of total traffic at the station with a particular destination
q=1–p
W = desired accuracy (percent error * p)
N = traffic volume at the survey station
Z = normal variate for specified level of confidence
 114

The total number of required completed surveys n is simply the sample rate r
times the traffic volume N.
While most reports on vehicle intercept and license plate follow-up surveys do
not explicitly state how the sample size was determined, many specify the sample rate at
which surveys were administered. For a specific confidence interval and error, the
sample rate increases as the true proportion p decreases. In addition, the sample rate
decreases as the traffic volume increases (the absolute number of samples increases,
however). This follows the results obtained from simulating the license plate matching
technique in the previous section in that, as the true proportion between two stations
decreases, the required capture rate increases (to estimate at the same level of
confidence and error). Therefore, the Hajeck’s equation is recommended for
determining sample sizes for vehicle intercept and license plate follow-up surveys.
The sample size n obtained from the Hajek is actually the required number of
completed surveys. Therefore, depending upon the method used, this number will have
to be increased to account for non-response of surveys and errors in the data collection
process.
For roadside interviews, a small percentage of drivers (generally <10%) will
refuse to participate in the survey. Therefore, the total number of vehicles stopped for
an interview should be n/0.9.
For the postcard questionnaire handout method, generally 15% - 30% response
can be expected. Prior to the actual study, tests should be conducted to more
accurately determine the response rate. Therefore, the number of postcards handed out
should be approximately n/0.15 (Virkud, 1995).
Finally, for the license plate follow-up survey technique, the number of license
plates recorded should again account for a 15% - 30% response rate, but also for errors
in recording license plates in the field (those that will not match with the BMV database)
which is assumed to be no greater than 10% based on the error analysis conducted in
Chapter 7. Therefore, the total number of license plates recorded should be
n/[0.15*0.90].

7.3 Vehicle Tracing Technique

While the vehicle tracing technique has not generally been used in practice on a
large scale, an analysis was conducted to determine the accuracy of the estimated OD
 115

matrix changes with varying degrees of sampling cell phones or GPS systems (probe
vehicles).

7.3.1 Simulation
This analysis assumes that the data obtained on probe vehicles was perfect,
meaning that the origin and destination zones in which a trip began and ended are
known.
To start, a simple 4x4 zone OD matrix was constructed, with each OD pair
containing 100 trips. Therefore, the total number of trips in the matrix was 1600. In
reality, these zones could represent actual traffic analysis zones or entry/exit nodes on
the cordon line of a study area. A vehicle identifier was created for each vehicle that
contained the origin and destination. From the 1600 vehicles, a sample of vehicles was
chosen randomly, from which the probe vehicle OD matrix was created. The probe
vehicle OD matrix was expanded globally to determine an estimated OD matrix. This
estimated OD matrix was compared to the true OD matrix by calculating the root mean
square error (RMSE) and percent root mean square error (PRMSE). Tables 18 – 20
illustrate this process.

Table 18: Actual OD Matrix (OD1)

OD1 1 2 3 4 Total
1 100 100 100 100 400
2 100 100 100 100 400
3 100 100 100 100 400
4 100 100 100 100 400
Total 400 400 400 400 1600

Table 19: Sample Probe Vehicle OD Matrix*

50% 1 2 3 4 Total
1 53 50 48 51 202
2 59 51 49 45 204
3 45 53 51 51 200
4 46 53 46 49 194
Total 203 207 194 196 800
*In this case, the probe vehicle matrix is based on a 50% sample.
 116

Table 20: Estimated OD Matrix (Expanded Probe Vehicle Matrix)

100%' 1 2 3 4 Total
1 106 100 96 102 404
2 118 102 98 90 408
3 90 106 102 102 400
4 92 106 92 98 388
Total 406 414 388 392 1600

The RMSE and PRMSE for the estimated OD matrix for Table 18 is 7.1 trips and
7.1%, respectively. The reason the PRMSE is calculated is to compare other OD
matrices that have a different total number of trips in the matrix. In addition, this matrix
was also sampled at different rates. Table 19 shows the probe vehicle OD matrix based
on a 50% sample (800 vehicles). The same matrix was estimated from probe vehicles
based on 1%, 5%, 25%, 50%, 75%, 95%, and 99% sample rates as well.
The process of obtaining the probe vehicle OD matrix was repeated 50 times for
each sampling rate. The mean of the RMSE and PRMSE were then calculated from the
50 simulations. Figure 54 illustrates the results of this analysis.

Mean PRMSE of Estimated OD M atrices

100%
Mean PRMSE of Estimated OD Matrices

90%

80%

70%

60%

50%

40%

30%

20%

10%

0%
0% 20% 40% 60% 80% 100%
Percent of Vehicles Sampled (Probe Sample)

Figure 54: PRMSE of Estimated OD Matrices based on OD1 (Table 18)

 117

Figure 54 illustrates that as the number of probe vehicles increase, the PRMSE
decreases, which is expected. At 50% sampling, the mean PRMSE drops to
approximately 10%.
Because this figure was developed only for a 16-cell matrix with an equal number
of trips per OD pair, it was desired to see how the error changed as the number of zones
of the true OD matrix changed or as the number of trips in a 16-cell matrix shifted so that
they were not uniformly distributed to each OD pair. This is discussed in the next
section.

7.3.2 Results
The same procedure was conducted on other types of OD matrices to determine
the relative change in the accuracy of the estimated OD matrix. Tables 21 – 24 illustrate
the other types of OD matrices that were randomly sampled.

Table 21: 4x4 1600-trip Matrix (OD2) without Uniformly-Distributed Cells

OD2 1 2 3 4 Total
1 0 0 0 0 0
2 7 7 7 7 28
3 49 49 49 49 196
4 344 344 344 344 1376
Total 400 400 400 400 1600

Table 22: 4x4 1600-trip Matrix (OD3) without Uniformly-Distributed Cells or Zeros
OD3 1 2 3 4 Total
1 7 7 7 7 28
2 7 7 7 7 28
3 42 42 42 42 168
4 344 344 344 344 1376
Total 400 400 400 400 1600

Table 23: 4x4 16,000-trip Matrix (OD4) with Uniformly-Distributed Cells

OD4 1 2 3 4 Total
1 1,000 1,000 1,000 1,000 4,000
2 1,000 1,000 1,000 1,000 4,000
3 1,000 1,000 1,000 1,000 4,000
4 1,000 1,000 1,000 1,000 4,000
Total 4,000 4,000 4,000 4,000 16,000
 118

Table 24: 8x8 1600-trip Matrix (OD5) with Uniformly-Distributed Cells

OD5 1 2 3 4 5 6 7 8 Total
1 25 25 25 25 25 25 25 25 200
2 25 25 25 25 25 25 25 25 200
3 25 25 25 25 25 25 25 25 200
4 25 25 25 25 25 25 25 25 200
5 25 25 25 25 25 25 25 25 200
6 25 25 25 25 25 25 25 25 200
7 25 25 25 25 25 25 25 25 200
8 25 25 25 25 25 25 25 25 200
Total 200 200 200 200 200 200 200 200 1600

After the probe vehicle matrix was created based on the varying degrees of
sampling and then expanded to estimate the original OD matrix for each of the true OD
matrices listed above, the mean PRMSE was calculated from 50 simulations and
compared to mean of each of the other OD matrices to see how the accuracy differed.
Figure 54 illustrates these results.

PRMSE of Estimated OD Matrices

100%

90%

80%

70%
OD1
60%
OD2
PRMSE

50% OD3

40% OD4
OD5
30%

20%

10%

0%
0% 20% 40% 60% 80% 100%
Probe Sample

Figure 55: Mean PRMSE for Various OD Matrices

It can be seen in Figure 54 that OD5 (Table 24) has the highest error, and OD4
(Table 23) has the lowest error for any given probe sample. In addition, OD1 (Table 18)
 119

and OD3 (Table 22) have equal error. How can this be explained? It seems that, given
a particular probe sample rate, the overall error for the OD matrix depends upon the
average number of trips per non-zero OD pair. For example, OD1 and OD3 both have
100 trips per non-zero pair. OD2 (Table 21), which has slightly more (133 trips per non-
zero pair), has slightly better overall accuracy at the same sampling rate. The same is
true regardless of the number of zones in the matrix. OD5, which has 25 trips per non-
zero pair, has far less accuracy. The opposite is true for OD4, which has 1,000 trips per
non-zero OD pair.
The problem in practice is that, if given a set of data from probe vehicles, the size
of the sample relative to the number of vehicles sought to be estimated in the OD matrix
(the sampling rate) is unknown. Without the sampling rate, it is difficult to expand the
probe vehicle matrix to the estimated OD matrix to obtain the number of trips per OD
pair. The probe sampling rate can be approximated by determining the average ratio of
traffic volumes from the sample to those of actual counts (such as tube counters) for a
number of links in the study area. This analysis, however, serves as a general starting
point for determining accuracy when estimating OD matrices from probe vehicle data
such as cell phones and in-vehicle GPS systems.
 120

CHAPTER 8 – GUIDELINES FOR SELECTING AN OD TECHNIQUE AND METHOD

The purpose of this chapter is to combine elements of the previous chapters, the
Travel Survey Manual, and other sources to assist a planner in determining the
technique and method by which an OD study should be conducted based on the
objectives of the study, cost constraints, desired accuracy of the results, and traffic
conditions in the study area.

8.1 Summary of Findings for License Plate Data Collection

Before identifying and recommending general guidelines, the following four
sections summarize the findings for license plate data collection as a result of the
analyses conducted as part of this project.

1. Vehicle Speed and Capture Rate

In Section 7.3, the maximum vehicle speed for identifying license plates
with half-sized (1.25”) characters was determined for manual recording methods
(clipboard, audio, and laptop) for various levels of visual acuity.
For example, a person with 20/20 visual acuity recording 4-characters off
license plates from vehicles traveling 30 mph would capture approximately 25%
of license plates with half-sized characters (Figure 39).
Because Indiana issues many license plates with characters that are
1.25” in height, the maximum speed at which approximately all license plates can
be captured is relatively low for a person with normal visual acuity (20 mph for 4-
character recording and 10 mph for full string recording). This maximum speed
would increase if only 2.75” characters were used on license plates
(approximately 45 mph for 4-character and 30 mph for full string recording).
 121

For these reasons, it is recommended that video be used for recording

license plates if the maximum vehicle speed is exceeded.

2. Vehicle Flow and Capture Rate

In Section 7.4, the maximum vehicle flow for recording license plates was
determined for the manual recording methods based on the amount of time
required to record a license plate and the headways between vehicles. For
example, if the audio method is being used on a roadway with uninterrupted flow
of 350 vehicles per hour per lane, the capture rate of those vehicles is ~85%.
Of the three manual methods, audio can achieve the highest capture rate
for a given flow rate. For flow rates above which capture rates cannot be
achieved by the clipboard, audio, or laptop methods, the video method is
recommended.
The values determined in Section 7.4 are for uninterrupted flow only. For
interrupted flow (e.g., downstream from a traffic signal), vehicle headways will be
much shorter (depending on the amount of queueing at the signal), so video is
always recommended.

3. Issues with Audio Recorders and Digital Video Camcorders

Chapter 6 discusses many of the issues when recording license plates
with audio or video equipment.
For audio, the primary issue is the quality of the recording (the clarity of
the license plate records during playback). Generally, it is recommended that the
highest recording quality be used. In addition, it was determined that cassette
audio recorders are better than digital because they have external storage
(tapes) that can be changed quickly and easily.
For video, there are many more issues to consider, such as set up
location, shooting angle, exposure settings, magnification, and shutter speed.
Section 6.8 describes all of these in greater detail.
When selecting video equipment, it is recommended that one camcorder
be purchased and that field tests be conducted prior to the actual study, in order
to ensure that a particular camcorder model is adequate for recording license
plates from the roadside.
 122

4. Accuracy of the License Plate Matching Technique

Section 8.1 evaluated scenarios of the license plate matching technique
in which the traffic volume, proportion between stations, and capture rates were
simulated at various levels in order to evaluate the accuracy (by calculating the
standard deviation of the percent error) of the estimated number of trips between
any two stations after 50 simulations.
Based on the results (Table 17), the accuracy of the estimate increases
as each of the variables increases (while holding the other variables constant).
For example, for a proportion of 5% and a capture rate product of 56.3%, the
standard deviation is 5.5% for a traffic volume of 5,000 vehicles, and 3.8% for a
traffic volume of 10,000 vehicles. Likewise, for a capture rate product of 90.3%
and a traffic volume of 5,000 vehicles, the standard deviation is 1.8% for a
proportion of 5% and 0.7% for a proportion of 25%. Finally, for a proportion of
25% and a traffic volume of 1,000 vehicles, the standard deviation is 9.7% for a
capture rate product of 25% and 4.8% for a capture rate product of 56.3%

8.2 Selecting an OD Technique

The first step in preparing to conduct an OD study is to select a technique. In
order to do this, the objectives of the study need to be considered, because different
techniques yield different information, and the objectives of the study will themselves
eliminate some of the techniques. The following paragraphs briefly describe the data
obtained by each technique.
In the license plate matching technique, only the number of trips between OD
pairs (observation stations) is collected, and the exact origins and destinations of the
trips are unknown. Socioeconomic data about each vehicle driver is not collected
because there is no interaction with the driver and no survey forms.
The license plate follow-up survey technique is a combination of various
techniques. Like the license plate matching technique, license plates are recorded at
various roadside stations. However, this technique also uses a mail-out/mail-in survey
to collect detailed trip (true origins and destinations) and socioeconomic information
about vehicle and driver.
 123

The vehicle intercept survey technique yields the same information as the license
plate follow-up survey technique. However, this technique requires direct interaction
with the driver, but also has much higher response rates. Therefore, this technique is
generally better suited for low - medium speeds and low flow situations.
The vehicle tracing technique is similar to the license plate matching technique in
that it collects strictly information on the number of vehicle trips between OD pairs. In
addition, exact origins and destinations are obtained. This technique, however, does not
yield socioeconomic data on the persons sampled. This technique is currently being
used by only a few agencies on a large scale, and the results are not yet known.

8.3 Selecting a Method

The method chosen for the license plate matching, license plate follow-up
survey, or the vehicle intercept survey techniques is largely based on the cost of each
method and the desired accuracy, both of which are also tied to the traffic conditions of
the road on which the data is being collected.

8.3.1 Accuracy
For the license plate matching technique, there are four basic methods for
collecting license plate data: clipboard, audio, laptop, and video. In Chapter 7, capture
rates for each of the manual methods (clipboard, audio, and laptop) were determined
based on vehicle speed and flow. In addition, the relative amount of field recording and
transcription error for each method was discussed. Chapter 6 evaluated the various
types of equipment and the issues associated with each when collecting license plate
data. Essentially, the video method should be used to achieve the most accurate results
at higher speeds and flows. However, if less accuracy is sufficient or video is cost-
prohibitive, manual methods can be used. The accuracy of the results using the license
plate matching technique was discussed in Chapter 8. Generally, as the study area
increases (in terms of geography, population, and number of external stations) the
proportions between stations decrease, which requires higher capture rates at each
station.
If the license plate follow-up survey technique is used, license plate data has to
be collected. As was done with the license plate matching technique, Chapter 7
evaluates the vehicle speeds and flows at which the video method should be chosen
 124

over manual methods for data collection. In addition, a survey has to be designed to
obtain the trip information, and a method for distributing and conducting the survey has
to be determined. While this was not evaluated as part of this project, the Travel Survey
Manual discusses these issues in greater length.
If the vehicle intercept survey technique is used, there are generally two practical
methods available: roadside interviews or postcard questionnaires. Because postcard
questionnaires require less time with the driver, this method should be used in low -
medium flow situations. However, because the interview method achieves a higher
response rate and better accuracy, the two methods may be used in combination to
avoid traffic delays. The number of interviews completed, however, is dependent on the
number of interviewers on site and the flow of traffic into and out of the interview site.
The safety of the interviewers and drivers and the flow of traffic should not be
compromised in an attempt to conduct more interviews.

8.3.2 Cost Items

Generally, it is difficult to accurately estimate study costs of various OD
techniques based on previous studies because of inflation, sampling rates (for survey
methods), and incomplete information provided by those studies. However, Table 25
highlights the general cost items associated with each OD study technique. This
information may be useful to determine the cost of a study or to compare the same study
using two different techniques.

Table 25: General Cost Items by OD Study Technique

License Plate License Plate Vehicle Intercept
Matching Follow-Up Survey Survey Vehicle Tracing
Study Design Study Design Study Design Study Design
Equipment Equipment Equipment Equipment
Observers Observers Personnel Probe Data
Training Training Training Data Reduction
Data Reduction BMV Fees Police Assistance
Survey Design Traffic Control
Survey Printing Survey Design
Survey Mailing Survey Printing
Data Reduction Survey Mailing
Data Reduction
 125

In the table above, “study design” refers to planning of the study (selecting the
technique and method and determining the sites and the amount of personnel required
at each site). “Equipment” refers to that required for field data collection (including
safety equipment, recording equipment, communications equipment, etc.). “Observers”
refer to the cost associated with the number of observers and the length of the study
period. “Training” refers to any instruction given to the study staff prior to the actual data
collection period. “Data reduction” refers to the amount of time required to transfer raw
data into a form that can be analyzed. There are other costs, such as travel and per
diem, that should be the same for all of the studies (except vehicle tracing).

8.4 Conclusions and Future Research

In conclusion, this report offers discussion and analysis of various data collection
techniques and methods for conducting Roadside Station Origin-Destination Studies. It
evaluated the different methods available for license plate data collection under various
traffic conditions. This report briefly analyzed the accuracy of the License Plate
Matching and Vehicle Tracing Techniques. Finally, it provided some general guidelines
for selecting the technique and data collection method depending on the objectives of
the study and traffic conditions of the roads in the study area.
In order to ensure the reliability of OD study results, better sample size formulas
should be developed, particularly for the license plate matching technique, that
incorporate traffic volume and capture rates at the stations and proportion of trips
between stations. As was shown in the Chapter 8 of this report, these variables may
cause the estimates to be inaccurate.
In addition, emerging technologies, such as vehicle tracing through wireless
phones, have the potential to provide data for a wide range of applications in
transportation beyond origin-destination data without time-consuming and costly field
data collection. However, there are still issues, such as privacy, cost, and data reliability
that need to be addressed in order for this technology to become a just, adaptable, and
efficient source of data.
 126

LIST OF REFERENCES

1. Asakura, Y., E. Hato, and M. Kashiwadani. Origin-Destination Matrices

Estimation Model Using Automatic Vehicle Identification Data and its
Application to the Han-Shin Expressway Network. Transportation. Kluwer
Academic Publishers, Netherlands, 2000, pp. 419-438.
2. Bonsall, P.W. and J. McKimm. Non-Response Bias in Roadside Mailback
Surveys. Traffic Engineering & Control, v34, n 12, London, Dec 1993, pp.
582-591.
3. Cellular Telecommunications & Internet Association.
http://www.ctia.org/research_statistics/index.cfm/AID/10030. July 2005.
4. Der Wann, C. and Y.M. Chen. “Mobile Location Tracking with Velocity
Estimation.” The IEEE 5th International Conference on Intelligent
Transportation Systems, pp. 566-571.
5. Easycamcorders.com. http://www.easycamcorders.com/content/Beginners-
Guide.htm. June 2005.
6. Federal Communications Commission. http://www.fcc.gov/911/enhanced/. May
2005.
7. Florida State Patrol,
http://www.fhp.state.fl.us/html/FHPInTheNews/StopSurvey.htm. April 2005.
8. Fricker, J.D. and R.K. Whitford. Fundamentals of Transportation Engineering; a
Multimodal Approach. Pearson Prentice Hall, Upper Saddle River, New
Jersey, 2004.
9. Gupta, R., J.D. Fricker, and D.P. Moffett. Reduction of Video License Plate Data.
Transportation Research Record 1804. TRB, National Research Council,
Washington, DC, 2002, pp. 31-38.
10. Hajek, J.J. Optimal Sample Size of Roadside Interview O-D Surveys. Ontario
Ministry of Transportation and Communication. Ontario, Canada, Jan. 1977.
 127

11. MacDonald, K. “Introduction to Video 101.” PULNiX America, Inc.

http://hexagon.physics.wisc.edu/research/technical%20info/pulnixvideoguide.
pdf, 2005.
12. Manual of Transportation Engineering Studies. Robertson, H.D, Editor, J.E.
Hummer, and D.C. Nelson. Institute of Transportation Engineers,
Washington, DC, 1994.
13. Martin, P.T., and M.C. Bell. Vehicle Tracking Through Unsampled Registration-
Plate Observation. Traffic Engineering and Control, v34, n 1, London, Jan
1993, pp. 8-12.
14. Nicholson, David. “Index of US License Plate Pages.”
http://www.15q.net/usindex.html. July 2005.
15. Oh, C. and S.G. Ritchie. Anonymous Vehicle Tracking for Real-Time Traffic
Surveillance and Performance on Signalized Arterials. Transportation
Research Record 1826. TRB, National Research Council, Washington, DC,
2003, pp. 37-44.
16. OnStar. http://www.onstar.com/us_english/jsp/index.jsp. May 2005.
17. Plotnikov, M. and P.W. Shuldiner. Effects of Ambient Light, Camcorder Settings,
and Automated License Plate Reader Settings on Plate Transcription Rates.
Transportation Research Record 1804. TRB, National Research Council,
Washington, DC, 2002, pp. 56-61.
18. Quiroga, C., R. Henk, and M. Jacobson. Innovative Data Collection Techniques
for Roadside Origin-Destination Surveys. Transportation Research Record
1719. TRB, National Research Council, Washington, DC, 2000, pp. 140-146.
19. Schaefer, M.C. “License Plate Matching Surveys: Practical Issues and Statistical
Considerations.” ITE Journal, Institute of Transportation Engineers,
Washington, DC, July 1988, pp. 37-42.
20. Shuldiner, P.W., S.A. D’Agostino, and J.B. Woodson. Determining Detailed
Origin-Destination and Travel Time Patterns Using Video and Machine Vision
License Plate Matching. Transportation Research Record 1551. TRB,
National Research Council, Washington, DC, 1996, pp. 8-17.
21. Slavik, M.M. Errors in Origin-Destination Surveys Done by Number-Plate
Techniques. Transportation Research Record 1050. TRB, National
Research Council, Washington, DC, 1985, pp. 51-58.
 128

22. Stokes, R.W. and T. Chira-Chivala. Design and Implementation of Intercity

Origin-Destination Surveys. Transportation Research Record 1236, TRB,
National Research Council, Washington, DC, 1989.
23. The Highways Agency, The Scottish Office Development Department, The Welsh
Office, and The Department of the Environment for Northern Ireland. “Traffic
Surveys by Roadside Interview.” http://www.archive2.official-
documents.co.uk/ document/deps/ha/dmrb/vol5/section1/ta1181a.pdf. United
Kingdom, 2005.
24. Travel Survey Manual. Cambridge Systematics, Inc. Travel Model Improvement
Program, USDOT Federal Highway Administration, Washington, DC, 1996.
25. Travel Time Data Collection Handbook. Turner, S.M., W.L. Eisele, R.J. Benz,
and D.J. Holdener. Texas Transportation Institute, College Station, TX, 1998.
26. Trimble. http://www.trimble.com/gps/index.html. July 2005.
27. Turner, S.M. Advanced Techniques for Travel Time Data Collection.
Transportation Research Record 1551. TRB, National Research Council,
Washington, DC, 1996, pp. 51-58.
28. Virkud, U. and C.S. Keyes. Design and Implementation of a Statewide Roadside
Origin-Destination Survey in Vermont. Transportation Research Record
1477. TRB, National Research Council, Washington, DC, 1995, pp. 15-25.
29. Wang, C.P. and S. Nallamothu. Basis of License Plate Recognition and a New
Approach. Congress on Computing in Civil Engineering, Proceedings, 1997,
pp. 143-152.
30. Washburn, S.S. and N.L. Nihan. Using Voice Recognition to Collect License
Plate Data for Travel Time Studies. Transportation Research Record 1593.
TRB, National Research Council, Washington, DC, 1997, pp. 41 – 46.
31. Welby, Marc J. “The License Plate Shack.” http://www.plateshack.com. July
2005.
32. Yahoo. “Missouri May Track Cell Phones for Traffic Data.”
http://news.yahoo.com/s/ap/monitoring_motorists, October 2005.