Ad Hoc Networks 11 (2013) 2115–2124

Contents lists available at SciVerse ScienceDirect

Ad Hoc Networks
journal homepage: www.elsevier.com/locate/adhoc

A study on traffic signal control at signalized intersections in

vehicular ad hoc networks
Hyeong-Jun Chang, Gwi-Tae Park ⇑
Department of Electrical Engineering, Korea University, Seoul, Republic of Korea

a r t i c l e i n f o a b s t r a c t

Article history: The Seoul metropolitan government has been operating a traffic signal control system with
Received 5 November 2011 the name of COSMOS (Cycle, Offset, Split MOdel for Seoul) since 2001. COSMOS analyzes
Received in revised form 16 February 2012 the degrees of saturation and congestion which are calculated by installing loop detectors.
Accepted 23 February 2012
At present, subterranean inductive loop detectors are generally used for detecting vehicles
Available online 7 March 2012
but their maintenance is inconvenient and costly. In addition, the estimated queue length
might be influenced by errors in measuring speed, because the detectors only consider the
Keywords:
speed of passing vehicles. Instead, we proposed a traffic signal control algorithm which
VANETs
Signalized intersections
enables smooth traffic flow at intersections. The proposed algorithm assigns vehicles to
Real-time control the group of each lane and calculates traffic volume and congestion degree using the traffic
Queue length information of each group through inter-vehicle communication in Vehicular Ad-hoc Net-
ITS works (VANETs). This does not require the installation of additional devices such as cam-
eras, sensors or image processing units. In this paper, the algorithm we suggest is verified
for AJWT (Average Junction Waiting Time) and TQL (Total Queue Length) under a single
intersection model based on the GLD (Green Light District) simulator. The results are better
than random control method and best-first control method. For a generalization of the real-
time control method with VANETs, this research suggests that the technology of traffic con-
trol in signalized intersections using wireless communication will be highly useful.
Ó 2012 Elsevier B.V. All rights reserved.

1. Introduction time-of-day (TOD), fixed-time control and real-time con-

trol methods. The time-of-day control method follows a
Nowadays, many countries are struggling with severe predefined signal timing plan by hour/day. The fixed-time
daily traffic congestion that causes a huge amount of social control method uses a signal timing plan set by an admin-
and economic loss. According to the report of Improve- istrator, while real-time control analyzes traffic informa-
ment of the Estimation Method for Traffic Congestion Costs tion acquired by sensors and builds a proper signal
from The Korean Transport Institute, the economic loss due timing control [2].
to traffic congestion in 2007 is estimated to be approxi- Time-of-day and fixed-time control methods have
mately $14.4 trillion [1]. Additional waste of time and en- advantages in the sense that they do not require additional
ergy are also a significant loss for individuals and nations. hardware and nor a complicated control algorithm. How-
To resolve such traffic congestion, traffic signal control ever, traffic congestion in modern urban areas is caused
methods are applied to improve traffic flow at intersec- not only by periodical rush-hours but also occasional
tions. The control methods can be largely classified into events interfering with the traffic flow, such as traffic acci-
dents and road construction. In addition, speed bumps,
curved roads and vehicle speed instability due to careless
⇑ Corresponding author. Tel.: +82 2 3290 3218; fax: +82 2 921 1325.
drivers can also cause traffic congestion. Therefore, time-
E-mail addresses: variety1@korea.ac.kr (H.-J. Chang), gtpark@
korea.ac.kr (G.-T. Park). of-day and fixed-time control methods can possibly even
URL: http://isrl.korea.ac.kr (G.-T. Park). increase traffic congestion instead [3].

1570-8705/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.adhoc.2012.02.013
2116 H.-J. Chang, G.-T. Park / Ad Hoc Networks 11 (2013) 2115–2124

Nomenclature

RQL queue length in the present cycle length MAXQ maximum queue length
QL final queue length for real-time signal control RCL temporal required cycle length
D identifier which is combined with path and NMAX  Q number of vehicles in the maximum queue
incoming direction length
VL vehicle length h vehicle-to-vehicle headway
ADBV average distance between vehicles GT green time
N number of group memebers GTT total green time
l length of lane RGT required green time
CL cycle length V/C flow ratio

On the contrary, the real-time control method is based cannot pass the intersection within red time and can be
on real-time sensing which potentially makes it an appro- used to determine whether green time needs to be ex-
priate strategy to resolve traffic congestion in modern ur- tended. It also allows controllers to clear the queue at the
ban cities. Moreover, recent improvement in converged intersection in order to improve traffic flow.
technologies of sensing and wireless networks has enabled The Seoul metropolitan government has been operating
the development of various real-time control methods. a traffic signal control system with the name of COSMOS
Attaining information of accurate vehicle detection is (Cycle, Offset, Split Model for Seoul) since 2001. COSMOS
the most important factor for real-time signal control. analyzes the degrees of saturation and congestion which
The most widely used sensors for vehicle detection at pres- are calculated by installing loop detectors such as forward
ent are spot traffic detectors and regional traffic detectors. detectors, left-turn detectors, spillback detectors and
Spot traffic detectors such as loop detectors and ultrasonic queue length detectors. Traffic control using queue length
detectors are sensors buried under the road, which makes is one of the most optimal real-time control methods. Its
their maintenance inconvenient and costly. Other types are disadvantage is that it requires detectors to be installed
microwave detectors and image detectors, which are easy in each lane since it relies on the data from both the up-
to install but also have high maintenance cost. The types stream queue length detector and the downstream spill-
of regional traffic detectors are AVI (Automatic Vehicle back detector. In addition, the estimated queue length
Identification), beacon and GPS (Global Positioning Sys- might be influenced by errors in measuring speed, because
tem) probe. Regional traffic detectors are generally high the detectors only consider the speed of passing vehicles.
priced and occasionally show low accuracy with regards To overcome such limitations, research with various sen-
to road conditions. In addition, both spot traffic detectors sors has been conducted. Some of the research proposed
and regional traffic detectors are only able to cover a lim- a method of obtaining local vehicle information using RFID
ited local area, and cannot used for route prediction [4]. tags attached to vehicles and RFID readers installed in each
In this study, a method of queue length estimation lane [5].
using communication between vehicles in a Vehicular Malik et al. [6] proposed a method to obtain traffic
Ad-hoc Networks (VANETs) environment is proposed. This information by installing sensor nodes in each lane and
method does not require the installation of additional controllers in each lane within a sensor network environ-
detectors and allows the estimation of optimal cycle length ment. Also, Khalil et al. [7] proposed a method to aggregate
and green split to enable real-time control of signalized vehicle information by installing pairs of arrival and depar-
intersections. ture nodes with one traffic signal server at intersections.
The remainder of the paper is organized as follows. Park et al. [8] proposed a queue length estimation model
that uses occupancy time to minimize errors caused by
 Traffic control system using VANETs (Section 3.1). the dependence on the average vehicle length and Instan-
 Intersection model and phase configuration. (Section taneous speed of the estimation process.
3.2) Jeong et al. [9] proposed a method that estimates de-
 Queue length estimation algorithm using VANETs (Sec- queuing time by measuring the delay time of individual
tion 3.3). vehicles before calculating the saturation flow ratio with
 Cycle length and green split estimation. (Section 3.4). the estimated de-queuing time. The estimated saturation
 Simulation environment (Section 4). flow ratio is used to calculate the queue length of each lane
 Simulation results (Section 5). and the final queue length is obtained after compensating
 Conclusions (Section 6). for errors. Lee and Oh [10] proposed a queue length esti-
mation algorithm using a pair of image detectors installed
2. Related works at upstream and downstream lanes.
This study proposes a real-time queue length estima-
In this chapter, research on vehicle queue length esti- tion algorithm that generates vehicle groups in each lane
mation and signal control system is described. Vehicle by using traffic signal cycle length and calculates the queue
queue length is defined as the number of vehicles that length using inter-vehicle communication within a
 H.-J. Chang, G.-T. Park / Ad Hoc Networks 11 (2013) 2115–2124 2117

VANETs environment. The proposed system enables more A traffic signal control algorithm which enables smooth
accurate queue length calculation than the existing loop traffic flow at intersections is proposed. This algorithm as-
detector methods, and it does not require additional hard- signs vehicles to a group within each lane and calculates
ware or complex calculations. A real-time traffic control traffic volume and congestion degree using traffic informa-
method using the estimated queue length is also proposed. tion of each group obtained via VANETs inter-vehicle com-
munication. It does not require the installation of
3. VANETs-based traffic signal control system additional devices such as cameras, sensors or image pro-
cessing units. Fig. 1 illustrates the proposed system.
3.1. System design
3.2. Single intersection control model and phase configuration
The proposed system is divided into following three
parts: (1) traffic information generator, (2) traffic signal In this study, a single intersection model for two arterial
controller, and (3) VANETs. The traffic information genera- roads with 2-lane crossing is used for real-time traffic con-
tor equipped in vehicles and traffic signal controller in- trol. Roads at the intersection are labeled according to their
stalled at intersections are onboard devices including direction as E (east), W (west), S (south) and N (north).
CPU, GPS modules, memory and a power source. It is as- Roads in each direction consist of two lanes, represented
sumed that a wireless VANETs environment can be used as L (left) and F (forward). Here, a right-turn is always al-
to enable vehicle-to-vehicle (V2V) and vehicle-to-infra- lowed. Hence, an intersection can be represented as a com-
structure (V2I) communication. bination of roads and lanes, and it is expressed as a

Fig. 1. Intersection traffic control using VANETs.

2118 H.-J. Chang, G.-T. Park / Ad Hoc Networks 11 (2013) 2115–2124

Fig. 4. Example of grouping vehicles.

the signal controller. Fig. 4 shows an example of grouping

vehicles.

3.3.1. Vehicle group generation

Groups are generated in each direction. There are four
paths marked as E (east), W (west), S (south) and N (north)
leading to the road intersection and each path has two
lanes in the incoming direction, which are left-turn (L)
Fig. 2. Single intersection with four ways. and go-forward (F). Each passing vehicle can thus have a
path PA of E, W, S, N and a direction DI of L, F, and each lane
where a vehicle is running can be determined by a pair of
direction indicator, D. D has the following eight compo-
PA, DI. As a result, there are at most eight lanes operating
nents. D = EF, EL, NF, NL, WF, WL, SF, SL. Each component
relative to the pair (PA, DI). This pair is represented by
of an intersection is illustrated in Fig. 2.
the direction identifier D. According to the direction iden-
For a real world experiment, the existence of pedestri-
tifier, each group leader is elected by the group members.
ans must be considered. Installing extra sensors to detect
An election period for the group leader starts at the begin-
pedestrians is beyond this study, so it is assumed that
ning of green time and depends on the dual ring phase con-
pedestrians always exist and a certain period of time is
figuration. A group leader periodically communicates with
allocated to take their crossing of the road into account.
group members during red time and calculates the group’s
Therefore a dual ring configuration is used for adequate
queue length. The calculated queue length is sent periodi-
phase control and pedestrians movement through the traf-
cally to the signal controller. The detailed vehicle grouping
fic flow in each lane. In this study, a lead forward dual-ring
method is as follows.
is used as shown in Fig. 3.
1. The signal at intersection is controlled by the phase
configuration in Fig. 3. After green time, vehicles stop-
3.3. Queue length estimation algorithm using VANETs ping at the intersection broadcast a group leader volun-
teer message to anonymous vehicles with the
For signal control at intersections, obtaining informa- maximum transmission power. Here, the message
tion on queue length information for each forward direc- includes a unique ID, time-stamp, location, velocity,
tion is crucial. Vehicles with wireless communication are route, intersection ID and the number of stops. The
able to transmit information directly to the signal control- vehicle that receives this message does not generate
ler. However, communication congestion might occur if its own message but instead re-broadcasts the received
there is a large number of vehicles in the queue. Also, the message.
concentration of data transmitted to the signal controller 2. The vehicle which issues the message compares its own
increases computation which may decrease the perfor- message to messages from other leader volunteer vehi-
mance of signal controller. The proposed method therefore cles, then sends a group leader yielding message to the
generates groups of vehicles in the same lane from the end vehicle which owns an earlier message issuing time and
of one green time to the beginning of the next, then elects is closer to the signal controller. Finally, the group lea-
group leaders who each transmit a group queue length to der is elected through this process.
3. The group leader broadcasts its election to its group
members, and vehicles that receive a completion mes-
sage of leader election send their information (ID, loca-
tion, speed, direction, vehicle size, leader ID, number of
stops and intersection ID) back to the group leader.
Once the group leader receives the message, it broad-
casts a group member acceptance message to the mem-
ber vehicle. After the member receives the acceptance
message, it updates its group leader’s ID and shares this
information with vehicles in its vicinity using periodic
Fig. 3. Lead forward dual-ring for green split. beacon messages.
 H.-J. Chang, G.-T. Park / Ad Hoc Networks 11 (2013) 2115–2124 2119

4. A group leader receives information from group mem- where RQL(t) represents the queue length in the present
bers during the whole cycle length excluding its own cycle length; D represents an identifier which is combined
green time and periodically transmits group queue with each path and incoming direction; VL represents the
length data to the signal controller. vehicle length; ADBV represents the average distance be-
5. When its own green time begins, the group leader tween vehicles; and N represents the number of group
accepts no more group member and passes through. members.
6. After the green time ends, steps 1–5 are repeated. A group leader sends group information to the signal
controller periodically. Here, the controller calculates the
The time for group generation and queue length estima- weighted average of the current and previous two queue
tion by signal cycle length is shown in Fig. 5. length values and uses the calculated result as the final
Each group is generated and the group leader elected queue length needed for real time signal control.
after green time. The elected leader then receives traffic
information from the group members and calculates the QLD ðtÞ ¼ A  RQLD ðtÞ þ B  RQLD ðt  1Þ þ C  RQLD ðt  2Þ
group queue length before the next green time. Group ð2Þ
leaders periodically transmit the group queue length to
the signal controller. where
At the beginning of green time, the group leader passes RQLD ðtÞ
through the intersection after transmitting the final group A þ B þ C ¼ 1; A ¼ ; B ¼ ð1  AÞ  A; C ¼ 1  ðA þ BÞ
lD
queue length to the signal controller. The signal controller
computes the next cycle length and green split time on the
and l represents a length of each lane
basis of the received group queue lengths when the green
As seen in Eq. (2) above, the weight is varied by the
time of phase 4 starts. After the green time of phase 4 ends,
number of vehicles in each lane. That is, if the number of
the traffic controller applies the new cycle length and
vehicles increases, the weight for the current queue length
green split time.
is also increased so that a prompt congestion control is
3.3.2. Queue length estimation possible. In contrast, if there are only few vehicles, the
A group is generated after a green time, and a group lea- weight for previous queue lengths increase, which enables
der receives vehicle information from its group members. smooth control without abrupt changes.
(Vehicle data include vehicle ID, time-stamp, location,
velocity, direction, group ID, intersection ID, number of
3.4. Cycle length and green split estimation
stops, and vehicle length.) The equation to estimate queue
length is as follows:
A signal control system must consider smooth traffic at
X
N the intersection and pedestrian safety as its highest prior-
RQLD ðtÞ ¼ VLi þ ADBV  ðN  1Þ ð1Þ ity. At the same time, it must meet the following
i¼0 objectives:

Fig. 5. Sequence of grouping vehicle and calculating queue length.

2120 H.-J. Chang, G.-T. Park / Ad Hoc Networks 11 (2013) 2115–2124

 minimize delays, Table 1

 minimize the number of stops, Cycle length estimation algorithm.

 maximize progression efficiency, Algorithm 1: Cycle length estimation algorithm(CLEA)

 minimize queue size at approaches, Input: QLi,i = 1, . . ., ,8
 maximize system throughput. Output: CL(t + 1)
1. Receive vehicle queue length from each group
All the control objectives listed above might not be met QL1(t),QL2(t), . . . , QL8(t)
2. Estimate the maximum queue length at each barrier
simultaneously under certain road and traffic conditions. A MAXQ1(t) = Max{QL1(t), QL5(t),QL2(t),QL6(t)}
delay is a measurement representing how much time a MAXQ2(t) = Max{QL3(t), QL7(t),QL4(t),QL8(t)}
vehicle spends traversing and stopping on the road. As a 3. Estimate the temporal required cycle length
method to minimize delay, decreasing the cycle length de- R CL1 ðtÞ ¼ N MAX  Q 1 ðtÞ  h
creases delays during red time. Once red time decreases, R CL2 ðtÞ ¼ N MAX  Q 2 ðtÞ  h
4. Compute a required cycle length
delays between cycles decrease too. However, a decreased
RCL(t) = RCL1(t) + RCL2(t)
cycle length also reduces green time so that the minimum 5. Estimate the required cycle length by adding intergreen time
number of stops increases in turn. For the optimal traffic RCL(t) = RCL(t) + intergreentime
signal control at intersections, cycle length and green split intergreentime = yellowtime + redtime
need to be configured appropriately based on an estima- 6. Determine the final cycle length compared with current cycle length
if{DC 6 (RCL(t)  CL(t)) 6 DC}then
tion of the queue length. This paper aims to provide mini- CL(t + 1) = CL(t)
mum delay time and minimum queue length. For this else
purpose, optimal cycle length and green split calculations CL(t + 1) = RCL(t)
are required. end if
7. Verify whether the computed cycle length satisfies mininum and
maximum conditions
3.4.1. Cycle length estimation algorithm
A cycle length needs to be long enough to expose the
critical movements in the longest lane at an intersection,
but if should not be exceedingly long. If a cycle length is Table 2
Green time estimation algorithm.
too short, phase changes occur too frequently and green
time is also shortened, which takes up more time than nec- Algorithm 2: Green time estimation algorithm(GTEA)
essary or desired. On the contrary, a great cycle length Input: CL(t + 1), RCLi, GTT, i = 1, 2
leads to increased waiting time and causes vehicles to wait Output: GTD
too long to be exposed from the intersection. Fig. 6 depicts 1. Estimate a required barrier green time using cycle length
the relationship between cycle length and delay. GT Barrier 1 ¼ ðR CLR1þR
CL1
 CL2 Þ
 CLðt þ 1Þ
In this paper, we propose a cycle length estimation GT Barrier 2 ¼ 1  GT Barrier1
2. Estimate a ratio of green time for each lane
algorithm based on vehicle queue length. A description of
R GT D ¼ PNV D =C D
the algorithm is shown in Table 1. D¼1
V D =C D

3. Choose a pattern in Table 3 and apply the ratio of green time to each lane
3.4.2. Green time estimation algorithm 4. Determine the final green time for each lane
GTD = RGTD  GTT
Once a cycle length is determined, a green time can be
derived by substracting an intergreen time (the sum of yel-
low and red time) from the time of a cycle length. From the
number of lanes (in this paper N = 8); and GTT represents
full green time, a green time ratio used for each phase is
the total green time.
described in (3).
V D =C D
GT D ¼ PN  GT T ð3Þ
D¼1 V D =C D

where GTD represents the green time at each lane; VD/CD

represents a flow ratio in each lane; N represents the

Fig. 6. Shape of a typical delay-versus-cycle length curve for an isolated

signal. Fig. 7. Single intersection model for simulation.
 H.-J. Chang, G.-T. Park / Ad Hoc Networks 11 (2013) 2115–2124 2121

Table 3
Green split combinations in lead forward dual-ring.

Fig. 8. Total waiting queue length.

2122 H.-J. Chang, G.-T. Park / Ad Hoc Networks 11 (2013) 2115–2124

Table 4 4.1. The structure of the intersection

Input flows for a single intersection.

Approaches Vehicle category Flow (vehicles/s) The simulations have been conducted to optimize
101 Sedan 1.00 queue length in a single intersection without considering
bus 0.25 the influence of adjacent intersections in Fig. 7.
102 Sedan 1.00
bus 0.25 4.2. Input vehicle generation
103 Sedan 0.25
bus 0.05 In the data generation process for the experiment, time
104 Sedan 0.25 intervals between arrivals are randomly generated for a
bus 0.05 single intersection, and service time is generated to follow
a single queuing model, M/M/1, which is an exponential
distribution, and the number of vehicle arrivals follows
Poisson distribution. For the experiment, a congested traf-
In this paper, we propose a green time estimation algo- fic during rush-hour is assumed and the number of enter-
rithm based on vehicle queue length. A description of the ing vehicles for the 4-way approaches at an intersection is
algorithm is shown in Table 2 and green split combinations shown in Table 4.
in lead forward dual-ring is shown in Table 3.

5. Performance evaluation
4. Experimental environment
The experiments have been carried out for enough time
To test the proposed algorithm, a Green Light District (2000 cycles) to analyze the congestion at an intersection
Simulator (GLD) was used [11]. GLD is an open-source and each experiment has been repeated 10 times. The re-
Java-based traffic simulator that enables road/intersection sult was compared with the following two algorithms to
design and allows the expansion of source codes to add evaluate the performance of the proposed algorithm. One
new algorithms for traffic signal control. New maps were is the random control and the other is best-first control.
generated and the source code was expanded to add the Best-first control always selects the traffic light configura-
algorithm proposed in this paper. For the inter-vehicle tion which sets the lights to green for the largest amount of
communication, a packet-based communication simulator vehicles in the lane [11].
such as NS-2 required, but a GLD simulator with added in- Random control is not able to consider the increasing
ter-vehicle communication can also be used since the pro- number of vehicles at an intersection so that it shows the
posed system utilizes only inter-vehicle communication in largest average waiting time and the largest waiting queue
VANETs environment. length. Since best-first control prioritizes the approaches

Fig. 9. Average junction waiting time with 95% confidence interval.

 H.-J. Chang, G.-T. Park / Ad Hoc Networks 11 (2013) 2115–2124 2123

101, 102 with a large number of vehicles, it gives a very lead to a complete re-grouping of the network. However,
long waiting time in the approaches 103, 104 with a small this problem does not arise in our approach. Because the
number of entering vehicles. Therefore, the total waiting group formation procedure is started when the vehicles
queue length of the proposed algorithm does not show a of the same direction are stopped near the intersections.
distinguishable difference from best-first control depicted Thus, the cost of group management caused by joining
in Fig. 8. This effect is offset because vehicles in a con- and leave operations of new vehicles is minimized.
gested lane have less waiting time while other vehicles Proposed algorithm makes use of control messages for
in not congested lanes have a longer waiting time. the group formation. Table 6 presents a summary of the
However, the best-first method has long and irregular communication parameters and their values.
waiting time on average as shown in Fig. 9. As can be seen Fig. 10 compares the overhead of the two communica-
from the graph, the confidence interval for the best-first tion method in bytes. As can be seen from the graph, the
control method is larger than other control methods. It overhead depends on the number of vehicles. In direct
means that the best-first method might not guarantee communication, all vehicles send their information period-
attainment of the stable control in congested traffic envi- ically to the signal controller. Therefore, the amount of
ronment. On the other hand, the proposed algorithm overhead in communication is proportional to the number
shows that it has stable waiting time on average at the of vehicles as well as the waiting time in the intersections.
junction. Table 5 shows the means and 95% confidence On the other hand, proposed communication method by
intervals of each control method. using grouping does not depend on the waiting time. Be-
Group formation in a vehicular ad hoc networks is con- cause the vehicle members send their information back
sidered as costly and difficult problem. In many ap- to the group leader only when they receive a completion
proaches, the overhead in communication increases with message of leader election.
the network dynamics, a single change in the group can
6. Conclusions
Table 5
Average junction waiting time – mean, 95% confidence interval. In this study, a real-time traffic control system on the
Control method Mean CI Lower CI Upper CI basis of VANETs is proposed. This system estimates the
queue lengths in each lane and determines cycle lengths
Random control 17.277 0.228 17.049 17.505
Best-first control 15.456 0.332 15.124 15.788 and green splits for a traffic signal controller. The perfor-
Proposed control 12.778 0.195 12.583 12.974 mance of the proposed algorithm is evaluated by conduct-
ing simulations compared to the existing control methods.
The result of this study can be summarized as follows.
First, an algorithm to estimate queue lengths for each
Table 6
Configuration values for inter-vehicle communication. lane based on inter-vehicle communication is proposed.
A group leader is elected and a group is generated for vehi-
Parameter Value
cles driving in the same direction according to the cycle
Transmission data rate 3 Mbit/s length of a traffic signal controller, and information from
One hop communication distance 250 m the generated group is sent to the controller. The controller
Packet size 100 bytes
Packet generation rate 500 ms
calculates the weighted average of the current and previ-
ous two queue length values and uses the calculated result

Fig. 10. Comparison of overhead.

2124 H.-J. Chang, G.-T. Park / Ad Hoc Networks 11 (2013) 2115–2124

as the final queue length needed for real time signal Journal of Information Science and Engineering 26 (3) (2010) 753–
768.
control.
[8] HyunSeok Park, YoungChan Kim, HakRyong Moon, A development of
Second, an algorithm to estimate cycle length and green traffic queue length estimation model using occupancy time per
split is proposed. A cycle length is calculated on the basis of vehicle based on COSMOS, Journal of Civil Engineering 27 (2D)
the estimated queue length. After the cycle length is calcu- (2007) 159–164.
[9] YoungJae Jeong, YoungChan Kim, HyunSoo Paek, Development of the
lated, a barrier length can be determined by estimating the signal control algorithm using travel time informations of sectional
time required for the four directions. A green time is as- detection systems, Journal of Korean Society of Transportation 24 (7)
signed in proportion to the required time for each (2006) 181–191.
.
[10] ChulGi Lee, YoungTae Oh, Development of the optimal signal control
direction. algorithm based queue length, Journal of Korean Society of
The proposed algorithm was compared to the random Transportation 20 (2) (2002) 135–148.
and best-first control methods. The total waiting queue [11] Wiering, M., Vreeken, J., van Veenen, J., Koopman, A., Simulation and
optimization of traffic in a city, IEEE Intelligent Vehicles Symposium,
length is shortened compared to random control, and the 14–17 June 2004, pp. 453–458.
proposed algorithm shows a minimized waiting time for
each individual vehicle when the green split is assigned HyeongJun Chang received his B.S. degrees in
for each direction in accordance with the amount of traffic Control and Instrumentation Engineering
from Korea University in 2005. He is currently
flow.
an Integrated Master and Ph.D. course student
at the School of Electrical Engineering in
Korea University. He is a member of the Kor-
References ean Institute of Electrical Engineers (KIEE), the
Institute of control, automation, and system
[1] HanSun Cho, InGi Park, DongMin Lee, JunSeok Park, Improvement of engineers Korea (ICASE). His research inter-
the Estimation Method for Traffic Congestion Costs, The korea ests include wireless sensor networks, vehic-
Transport Institute Research Report 2007-07, Gyeonggi-do, Korea, ular ad hoc networks, and nonlinear control.
July 2007, pp. 1–219.
[2] R. Gordon, R. Reiss, H. Haenel, E. Case, R. French, A. Mohaddes, R.
Wolcott, Traffic Control Systems Handbook, Department of GwiTae Park received his B.S., M.S. and Ph.D.
Transportation, Federal Highway Administration, Washington, DC,
degrees in Electrical Engineering from Korea
2005.
University in 1975, 1977 and 1981, respec-
[3] M. Artimy, Local density estimation and dynamic transmission-
tively. He was a technical staff member in the
range assignment in vehicular ad hoc networks, IEEE Transactions on
Intelligent Transportation Systems 8 (3) (2007) 400–412. Korea Nuclear Power Laboratory and an elec-
[4] Lawrence A. Klein, Milton K. Mills, David R.P. Gibson, Traffic Detector trical engineering faculty member at Kwang-
Handbook, third ed., vol. I, US Department of Transportation Federal Woon University, in 1975 and 1978,
Highway Administration, Georgetown Pike, 2007. respectively. He joined Korea University in
[5] GangDo Seo, Distributed traffic signal control using prediction model 1981 where he is currently a Professor in
of intersection queue length, Ph.D. Thesis, School of Electronics school of Electrical Engineering. He was a
Engineering College of it Engineering, Kyungpook University, 2009. Visiting Professor at the University of Illinois
[6] Tubaishat Malik, Shang Yi, Shi Hongchi, Adaptive Traffic Light in 1984. He is a fellow of the Korean Institute
Control with Wireless Sensor Networks, Consumer of Electrical Engineers (KIEE), the Institute of control, automation, and
Communications and Networking Conference, 2007. CCNC 2007. system engineers Korea (ICASE), and advisor of Korea Robotic Society. He
4th IEEE, January 2007, pp.187–191. is also a member of Institute of Electrical and Electronics Engineers (IEEE),
[7] Khalil M. Yousef, Mamal N. Al-Karaki, Ali M. Shatnawi, Intelligent
Korea Fuzzy Logic and Intelligent Systems Society (KFIS).
traffic light flow control system using wireless sensors networks,