Ain Shams Engineering Journal 14 (2023) 102115

Contents lists available at ScienceDirect

Ain Shams Engineering Journal

journal homepage: www.sciencedirect.com

Traffic safety assessment for roundabout intersections using drone

photography and conflict technique
Mohamed Shawky a,⇑, Alsayed Alsobky a, Ahmed Al Sobky b, Ahmed Hassan c
a
Department of Public Works, Ain Shams University, Cairo, Egypt
b
Department of Full Stack Development, Nilecom, Cairo, Egypt
c
Department of Public Works, Cairo University, Giza, Egypt

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

Article history: Road design deficiencies and improper driver behavior at roundabout intersections may result in traffic
Received 25 October 2022 bottlenecks, irregular traffic patterns, and potential crashes. Thus, road safety inspection is conducted to
Revised 26 December 2022 identify potential safety hazards and propose safety measures. The traditional safety inspection depends
Accepted 30 December 2022
on unreliable traffic collision data visual data collection and superficial analysis. In this regard, surrogate
Available online 18 January 2023
safety assessment approaches are utilized to overcome the limitations found in traditional approaches.
This paper employs an innovative surrogate approach for such a process by analyzing videos captured
Keywords:
by a drone. A video processing technique is applied to determine the vehicle trajectories and extract con-
Traffic conflicts
Traffic safety assessment
flict points. Accordingly, the conflict data are analyzed in terms of location, direction, and post-
Video processing encroachment time (PET) as a safety measure to identify potential safety problems related to intersection
Conflict measures geometry and driver behavior. This methodology is applied to an intersection case study in New Cairo
Roundabout intersection City, Egypt. The findings of this study confirm the interaction between intersection geometry, drivers’
behavior, and road safety on the examined safety measures.
Ó 2023 THE AUTHORS. Published by Elsevier BV on behalf of Faculty of Engineering, Ain Shams Uni-
versity. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/
by-nc-nd/4.0/).

1. Introduction among the sustainable development goals (target 3.6) of the UN-
2030 agenda [2].
With the growing number of smart cities and corresponding Analysis of reported traffic crash data is a staple of the conven-
advanced mobility requirements, the need to apply unconven- tional approach to evaluating and enhancing road safety. However,
tional methods to assess the safety of road users increases. Around especially in low and middle-income countries, crash data have
1.35 million people die annually and more than 50 million suffer several problems in many counties regarding their availability,
injuries as a result of road traffic crashes (RTCs) [1]. World Health accuracy, and adequacy. Thus, as an alternative and reliable road
Organization (WHO) reported that the majority (more than 90 %) of safety assessment approach, researchers have long relied on using
RTCs occur in low- and middle-income countries to become the surrogate safety measures (SSMs) such as the temporal proximity
leading cause of death for people aged 5–29 years old [1]. There- conflict technique including time-to-collision (TTC), and post-
fore, the United Nations (UN) has declared traffic road safety encroachment time (PET) measures which are widely utilized.
The concept of traffic conflict was first proposed by Perkins and
Harris in 1967 [3]. They defined traffic conflict as any potential
⇑ Corresponding author. crash situation leading to the occurrence of evasive actions such
E-mail addresses: m_shawky@eng.asu.edu.eg (M. Shawky), assobky@eng.asu. as swerving or braking. Recent research demonstrated the accuracy
edu.eg (A. Alsobky), assobky4@outlook.com (A. Al Sobky), ahmedehassan@gmail. and reliability of using traffic conflict analysis to estimate crash
com (A. Hassan). frequency and severity levels [4–6]. Such shifting from the reactive
Peer review under responsibility of Ain Shams University. concept (crash data-based) to the proactive concept (conflict data-
based) plays a dominant role in improving road safety, especially in
the forthcoming era of Connected and Autonomous Vehicles
(CAVs) [7–9]. In addition, traffic conflict analysis is a potent
Production and hosting by Elsevier method to examine the contributing factors affecting road safety

https://doi.org/10.1016/j.asej.2023.102115
2090-4479/Ó 2023 THE AUTHORS. Published by Elsevier BV on behalf of Faculty of Engineering, Ain Shams University.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
M. Shawky, A. Alsobky, A. Al Sobky et al. Ain Shams Engineering Journal 14 (2023) 102115

when crash data reports are missing or significant information is models recognized more hazard conflicts than the corresponding
absent [10]. In this approach, no need to wait for more road crashes univariate models.
and fatalities to happen before taking the appropriate actions and The types of traffic conflicts are generally classified in literature
countermeasures. into three types based on the measured angle between the two
Emerging new technologies of artificial intelligence (e.g., machine conflicting vehicles; rear-end collision for angles in a range of 0
learning) in tracking vehicles enhance the extraction of accurate traf- to 30°; lateral collision by lane change from 30 to 85°; and by a
fic conflict measures to assess road safety either online or offline [11]. crossing of frontal from 85 to 180° [25,26]. In general, there is an
Thus, SSMs are becoming increasingly important for the future of agreement that the threshold value of 5.0 s for PET is the best value
road safety analysis utilizing advanced technologies such as video to distinguish between conflict and non-conflict events [25–28].
recording by Unmanned Arial Vehicles (UAV), commonly called Regarding the threshold values of the PET, several studies
drones, image processing, and vehicle tracking programs. attempted to define the appropriate thresholds yet, the critical
As in most low and middle-income countries, the behavior of threshold value is an open question. The FWHA recommended
Egyptian drivers significantly affects road safety due to the lack threshold values for traffic conflict studies from 0 to 5 s for PET
of cultural understanding of traffic safety aspects [12,13]. This and 0 to 1.5 s for TTC [26]. In addition, a PET value of 2.5 s is
could be particularly recognized at unsignalized intersections severely used as a threshold between serious and non-serious con-
where the absence of priority rules and lane configuration leads flicts, while the threshold value of 1.0 s is strongly correlated with
to more improper and serious conflicts. However, up to date, there severe crashes (i.e., crashes result in injuries or fatalities [15,16,31].
is no significant attention to understanding how such drivers
behave to avoid collisions at unsignalized intersections, especially
3. Data collection and methodology
in the case of unmarked large roundabouts. Consequently, a funda-
mental problem could arise and would need prompt and efficient
3.1. Vehicle detection and tracking process
solutions. Therefore, this study employed the traffic conflict con-
cept to assess road safety and investigate drivers’ behavior at
Vehicle detection is a computer vision technique where objects
roundabout intersections by using drone video records in Egypt.
are located and classified in images or videos. The object detection
algorithm applies either machine learning or deep learning. This
algorithm aims at replicating human capabilities in recognizing
2. Literature REVIEW
objects [32]. Moreover, vehicle tracking is the process of detecting
objects through different video frames such that the position and
The time-to-collision (TTC) and post-encroachment time (PET)
direction are identified at each frame [33]. Then, the object’s trajec-
are the most traffic conflict measures in road safety studies as sur-
tory could be developed. Thus, these techniques are crucial to var-
rogate safety measures [14]. TTC is defined as ‘‘The projected time
ious applications such as traffic monitoring, management, etc.
until two interacted road users would collide if they continued on
Vehicle tracking is applied to extract their trajectories while cross-
their collision course with unchanged speeds and direction” [8].
ing the case study roundabout intersection which is located in New
Whereas PET is defined as ‘‘The elapsed time between the depar-
Cairo, Egypt at the location of Longitude 31° 280 2700 E and latitude
ture of an encroaching vehicle and the arrival of a trailing vehicle
30° 40 1500 N. A fifty-minute video was shot by a drone and is split
at the same point” [9]. The strong relationship between traffic con-
into segments, each two-minutes long. At each video segment, a
flicts and crashes has been confirmed in several studies. For
sample of vehicles is selected for the tracking process; the sample
instance, Anarkooli, et al. [15] showed a clear linear relationship
includes 1,217 vehicles which represents around 30 % of the vehi-
between traffic conflict and crashes (R2 = 0.85 for all crashes and
cles traversing the function area of the selected roundabout. Then,
R2 = 0.94 for fatal/injury crashes). Peesapati et al. [16] proved that
at each video frame, each selected vehicle position is reported to
the correlation between PET value of 2.0 s and crashes is 0.17 but at
develop vehicle trajectories across the intersection.
1.0 s the correlation reached 0.61.
Next, object tracking includes two main levels which are Single
in addition, Zheng et al. [17] conducted a thorough review of
Object Tracking (SOT) and Multiple Object Tracking (MOT) [34].
prior traffic safety studies that employed traffic conflict as a safety
The MOT approach aims to develop the trajectory for multiple
measure. Their findings demonstrated the distinctive feature of
objects in a video. The MOT starts with localizing the objects in
using the traffic conflict technique is its multi-dimensionality,
each frame with bounding boxes. Then, relating the objects among
issues with short observation periods, and temporal and spatial
different frames [34]. Two main techniques are developed for
correlation. Another review paper [18] raised critical questions
object tracking: visual feature tracking and classification neural
before the expected widely deployed autonomous vehicles such
network [35]. The visual feature tracking includes several algo-
as ‘‘Is a single SSM good enough for automated connected vehicles?
rithms such as Boosting, MIL, GOTURN, MOSSE, KFC, CSTR (DCF-
‘‘What SSM is better for trajectory optimization to improve safe-
CSR), Median flow, and TLD trackers that have been developed [35].
ty?”. The study showed that traffic conflict modeling and analysis
In the current research, the DCF-CSR algorithm is applied to
still require further investigation. Arun, et al., [19] conducted a sys-
track the vehicles which provide excellent performance on all stan-
tematic review of 386 studies on traffic conflict techniques. They
dard benchmarks in short-term object tracking [36]. The algorithm
summarized the gaps in the reviewed studies in four points; the
constructs a spatial reliability map that is used to adjust the filter
lack of suitable techniques to estimate crash risk by severity levels;
to focus on the part of the selected frame (i.e., Region of Interest).
the primary focus on signalized intersections, and the roundabout
Thus, the accuracy of the localization of the selected vehicle is
studies represented only about 9 %; the lack of suitable conflict
improved [37]. For each video segment, several vehicles are
measures for vulnerable road users; and the scarcity of validation
selected for tracking as shown in Fig. 1a. Then, Fig. 1b shows the
studies for conflict measures. Recently, a significant research effort
final recorded trajectories.
was done to use multi-conflict measures for road safety and apply
this approach to real-time safety assessment modeling [20–22]. Fu
and Sayed [23,24] provided multivariate modeling methods for 3.2. Treatment of drone instability
road safety assessment based on conflict extremes in real-time sit-
uations. They applied a mix of three conflicts indicated at signal- Object tracking using drone videos is a challenging process as it
ized intersections where the results showed that the multivariate captures the objects while UAV is not exactly at the same location
2
M. Shawky, A. Alsobky, A. Al Sobky et al. Ain Shams Engineering Journal 14 (2023) 102115

Vehicle detection recorded t

Fig. 1. Vehicle detection examples and trajectories recorded for the tracked vehicles.

resulting in distortion, occlusion, change in illumination, and blur- each frame separately. The five control points are selected in the
ring [38–40]. Therefore, there are significant differences in the condition of covering four corners and the center of the round-
resulting coordinates even for static objects (e.g., landmarks), using about intersection. These control points are then used to adjust
various video frames. the location of the same control points across the video frames.
In particular, the location error of the same landmark (i.e., static The root means square of errors reached 3.97 pixels (i.e.,
object) in different frames compared to the first frame due to the 0.97 m); while, the average was 1.77 pixels (i.e., 0.43 m) which
drone instability in this case study reaches 58 pixels (i.e., 14.1 m) ensures the high accuracy of instability treatment. Fig. 2 illustrates
as a root means square of errors with an average error 15 pixels the used control points (i.e., static landmarks) which are repre-
(i.e., 3.7 m). These errors ensure the importance of drone instability sented by the centers of green circles. The figure also shows the dif-
treatment to obtain accurate locations and then reliable safety ferent treated locations for these control points in all frames which
measures. are represented by the red points.
Often, these deformations are mainly classified into translation
and rotation. The translation is also divided into two types: hori- 3.3. Conflict points identification
zontal and vertical. The vertical translation causes scaling deforma-
tion, while the horizontal translation causes shifting deformation. The conflict point is identified using the PET indicator. The con-
In addition, rotation is divided into two types: rotation about the flict point is located at the intersection between vehicle trajectories
vertical axis and rotation about a non-vertical axis which causes where the PET value is 5 s or less. Then, the severity and type of
tilting deformation. conflict are defined based on PET ranges and the conflict angles,
Authors attempted to combine and treate all these deforma- respectively as given in Table 1. The conflict point is identified if
tions (i.e., shifting, horizontal rotation, scaling, and tilting) using its PET value is within a range of [0, 5]. The PET is calculated
only one equation as illustrated in Equation (1). The coordinates between two conflicting vehicle trajectories as shown in Fig. 3.
(xij , yij ) of an object (i) in a certain video frame (j) are projected into Then, to define the conflict severity, this range is divided into four
the coordinate system (X iR , Y iR ) in the reference frame (R) (e.g., the segments: Extreme, serious, Moderate, and General based on
first frame in the video). ranges of PET of [0,1], [1,2.5], [2.5, 3.5], and [3.5,5], respectively.
Also, the conflict is classified by type to rear-end, lane change,
X iR Sxj coshj Sxj sinhj dxj xij
1 and crossing based on the angle between conflicting vehicle trajec-
½ Y iR  ¼ ½ Syj sinh Syj coshj dyj ½ yij  ð1Þ
t 1j xij þ t 2j yij þ 1 tories at the conflict point. Table 1 provides the angle ranges for
1 t 1j t2j 1 1 each type based on SSAM developed by FWHA [26]. The case study
Where the transformation parameters for a frame (j) are: intersection is divided into sections that are selected to include dif-
ferent conflicts within the circulating lanes, approaches, and exits,
ddxj and dyj : the horizontal shift in directions of X-axis and Y-
where the intersection is divided into 12 sections, as indicated in
axis,
Fig. 4. The circulating lanes are divided into 4 sections. In addition,
hj : the horizontal rotation about Z  axis,
a section is defined for each approach and exit. At each circulating
Sxj and Syj : the scaling in directions of X  axis and Y  axis due
section, the conflicts rate is calculated by dividing the number of
to vertical translation,
conflicts by the exposure as given in Equation (2). In this section,
t1j and t2j : the effects of tilt deformation in directions of X  axis
the exposure is calculated using the conflicting volumes that are
and Y  axis.
approaching and circulating traffic volumes as given in Equation
To obtain these transformation parameters through the differ-
(3).
ent video frames, five well-distributed control points are identified
in all frames including the reference frame. Therefore, the transfor- total number of conflicts
Conflictrate ¼ ð2Þ
mation parameters in Equation (1) (i.e., dxj , dyj , hj , Sxj , Syj , t 1j and t2j ) Exposure
for each frame are determined using nonlinear optimization to
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
transform the coordinates of the control points in each frame to Approaching Volume  Circulating Volume
match as much as possible the corresponding coordinates in the Exposure ¼ ð3Þ
1000
reference frame. Accordingly, the transformation parameters for
a frame are used to obtain the adjusted coordinates of all detected
objects in such frame to the coordinates system of the reference 4. Analysis and results
frame.
This treatment is applied on all frames in the video (i.e., 104,769 The tracked vehicle resulted in 2,424 traffic conflicts at the cir-
frames, 30 frames per second) and then the corresponding trans- culating, entry, and exit sections. These conflict points are divided
formation parameters are determined by solving Equation (1) for into three types: rear-end, lane change, and crossing. The rear-end
3
M. Shawky, A. Alsobky, A. Al Sobky et al. Ain Shams Engineering Journal 14 (2023) 102115

Fig. 2. Used Control Points for Instability Treatment.

Table 1
Conflict point classification by type and severity.

Traffic conflict type Severity Angle

PET  1.0 s 1.0–2.5 s 3.5–5.0 s 3.5–5.0 s

Crossing Extreme Severe Moderate General kConflictanglek > 85
 
Lane change 85  kConflictanglek  30

Rear-end kConflictanglek < 30

Table 2
Analysis of conflict PET values in different sections.

Section Number of conflicts Average of PET Min. of PET Max. of PET Std. Dev of PET Range
Circular sections 1 303 3.0 0.1 5.0 1.2 4.9
2 264 2.8 0.0 5.0 1.3 5.0
3 663 3.0 0.0 5.0 1.2 5.0
4 805 2.9 0.5 5.0 1.2 4.5
Entry 5 13 2.0 0.4 4.8 1.4 4.4
6 171 3.0 1.2 5.0 1.1 3.8
7 45 2.5 0.0 4.8 1.4 4.8
8 2 3.8 2.8 4.8 1.4 2.0
Exit 9 8 4.2 1.6 4.9 1.1 3.3
10 97 2.9 0.6 4.9 1.2 4.3
11 16 2.6 0.1 4.9 1.4 4.8
12 37 3.1 0.8 4.8 1.1 4.0

conflicts represent about 13.2 % of the identified conflicts, while tion of the identified conflict points. The figure shows that sections
the lane change conflicts represent about 86.7 % of the conflicts. no. 3, and 4 are the most critical sections in terms of the total num-
Crossing conflicts are rarely found in the intersection. Fig. 5 illus- ber of identified conflicts as they include around 60 % of the total
trates the spatial distribution of the identified conflict points. The identified conflicts. In addition, sections no. 2, 3, and 7 include con-
figure shows that sections no. 1, 3, and 4 are the most critical sec- flicts with a PET value of zero indicating real crashes. The average
tions in terms of the total number of identified conflicts. of the PET value in entry sections 5 and 7 are the most critical. Fur-
thermore, section 6 is the most critical entry section in terms of the
number of identified conflict points. The conflict rate, calculated
4.1. Conflict analysis
based on Equations (1) and (2), is given in Table 3. The table shows
that section 4 is the most critical section with a rate of 3,174 con-
Conflict points are identified in terms of location, severity, and
flicts per thousand vehicles.
type. The location of the conflict points highlights the hazard level
of the section providing insights into the potential problems
4.1.1.2. Distribution in terms of conflict angle. Generally, roundabout
related to driver behavior, intersection geometry, etc.
intersections are designed to allow the drivers to merge into circu-
lar lanes with minimum angle values. However, based on the iden-
4.1.1. Conflict spatial distribution tified conflict points, the average conflicting angle in section 3 is
4.1.1.1. Distribution in terms of number and type. Based on the loca- the most critical with a value of around 22° as given in Table 4.
tion of the identified conflicts, Fig. 4 illustrates the spatial distribu- Also, the maximum conflicting angle in section 3 exceeds 90°
4
M. Shawky, A. Alsobky, A. Al Sobky et al. Ain Shams Engineering Journal 14 (2023) 102115

section 5 is an entry section; that may highlight a driving behavior

problem.
Furthermore, Fig. 6 shows a heatmap that presents an average
of conflicting points angle and PET values for different cells in
the intersections with a size of 1.5 m  1.5 m. The figure shows
the distribution of the severity of conflicts in the function area of
the intersection. From the figure, the circulating sections have
many serious conflicts. These high-severity conflicts could be
formed between approaching and circulating vehicles. The
approaching vehicles merge the circular sections at the most left
lane resulting in severe conflict with circulating vehicles. Also,
the figure indicates the severity of the conflict points at
approaches. Nonetheless, exits have significantly less severe con-
flict except in section 10 where U-turn maneuvers take place
resulting in serious conflict points.

Rear-end conflicts

Fig. 3. Definition of the PET measure for a conflict point [14]. Table 6 shows the share of rear-end conflict type in the
recorded conflict points. From the table, rear-end conflicts have
which implies occurring of crossing conflicts. Entry and exit sec- the major share in the recorded conflicts by around 87 %. Further-
tions have relatively small conflict angles ranging between zero more, there are about 38 %, 22 %, 12 %, and 8 % of these types of
to 5°. conflicts are located in sections 4, 3, 2, and 1, respectively. Thus,
the majority of these conflict types are found in the circulating
lanes where conflicts take place among approaching and circulat-
4.1.2. Conflict severity analysis ing vehicles, on the one hand, and also among circulating vehicles
The severity analysis is carried out based on the PET measure. as well, on the other.
Table 5 shows that the extreme conflict points represent about This type of conflict may be formed due to unsafe tailgating
3 %; while the serious conflicts are about 38 %. These values high- driving behavior where drivers may try to get close to the front
light a safety problem in the intersection. The safety problem may vehicle to avoid possible interference from other aggressive drivers
be related to driving behavior, geometric design, vehicle, or a com- who do not properly comply with the priority rules of the round-
bination of them. For instance, in section 10, there are serious rear- about intersections.
end conflicts. Such conflicts may be due to braking action carried
out by a vehicle in the main traffic stream to avoid collision with Lane change conflicts
vehicles performing unsafe merging maneuvers to the traffic
stream coming from section 6 using the U-turn. This problem Table 6 shows the share of lane change conflict type at circulat-
may be classified as driving behavior and improper geometric ing sections. The highest share is in section 3 with 62 % of lane
design of the intersection. changes recorded conflicts. This value indicates that the section
Table 2 shows the average PET value which indicates the sever- includes serious lane change conflicts due to driving behavior
ity level of each considered section. Section 5 has the least average and road geometric aspects.
PET value of 2.0 which is an indicator of serious conflict. However,

Fig. 4. Intersection segmentation.

5
M. Shawky, A. Alsobky, A. Al Sobky et al. Ain Shams Engineering Journal 14 (2023) 102115

Fig. 5. Spatial distribution of conflict points types at different sections.

Table 3 (AASHTO [44]) affects traffic conflicts and thus overall safety. The
Conflict rate at different sections of the circulating carriageway. geometric elements addressed include lane balance, entry and exit
Section No. of Conflict Angle PET Conflict rate angles, shifting of legs centerlines, and lane width.
1 303 19.5° 3.0 1,421
2 264 6.8° 2.8 1,570 4.2.1. Entry and exit angles
3 663 22.4° 3.0 1,913 The entry and exit angles play a significant role in directing dri-
4 805 6.3° 2.9 3,174 vers through the circulating carriageway and the conflicting angles
during lane change maneuvers. Fig. 7a and 7b show the typical
standard roundabouts with the acceptable paths’ movements.
Table 4 However, the actual situation of the intersection case study shows
Analysis of conflict angle values at different sections.
a significant shift in the two legs of the intersection (leg A and leg
Section Average of Min of Max of Std Dev of Range C) from the central island center. Such shifts caused an improper
Angle Angle Angle Angle conflict angle at sections 1 and 3, as proved by the conflict analysis
Circular 1 19.5° 0.3 62.4 19.5 62.1 shown in Table 4. This means that shifting a centerline of an inter-
2 6.8° 0.0 43.3 6.1 43.3 section leg results in a critical traffic safety situation at the circulat-
3 22.4° 0.3 91.9 24.5 91.6
ing sections of this leg entry.
4 6.3° 0.2 43.4 6.4 43.2
Entry 5 3.3° 0.5 6.1 2.0 5.7
6 5.5° 0.2 60.5 5.8 60.3 4.2.2. Lane-related issues
7 5.9° 0.3 17.4 4.0 17.0 Lane balance through interaction, the total number of circulat-
8 3.6° 3.3 3.8 0.4 0.5 ing lanes, and lane width are critical design elements in our case
Exit 9 2.3° 0.3 4.3 1.5 4.0
study. As shown in Fig. 8a there is an imbalance through the entry,
10 6.9° 0.4 27.8 5.8 27.3
11 2.1° 0.3 8.1 2.1 7.8 circular, and exit sections whereas the number of lanes ranges
12 4.2° 0.5 14.5 2.8 14.0 from 4 lanes to 7 lanes. This imbalanced situation in addition to
the large number of circulating lanes and island radius are usually
stated as non-preferable in the textbooks and now from this study
Table 5
proved that it results in dead areas, Fig. 8b, encouraging drivers to
Analysis of conflict severity. overlapping lanes that could be proved by vehicles trajectories;
examples in Fig. 8c and 8d.
Severity levels PET (sec.) f %f
Extreme 1 66 2.7 %
4.3. Driving behavior
Serious 2.5 918 37.9 %
Moderate 3.5 595 24.5 %
General 5 845 34.9 % Studying driving behavior is a vital part of traffic safety due to
the considerable share of human errors in traffic crashes. The driv-
ing behavior includes the actions of steering the vehicle such that
the traffic conditions may be affected by driver behavior. In this
4.2. Geometric design impact on conflicts regard, aggressive driving may create irregular traffic conditions
resulting in traffic crashes [41]. Aggressive driving includes over-
This section addresses how the deviation of the geometric speeding, tailgating, ignoring traffic regulations, improper lane
design from the standard rules of the roundabout intersection changing or weaving, etc. [42]. Fig. 9 shows a dangerous weaving
6
M. Shawky, A. Alsobky, A. Al Sobky et al. Ain Shams Engineering Journal 14 (2023) 102115

Fig. 6. Traffic conflict angle and PET value heat maps.

Table 6 4.3.1. Comply with priority rules

Share of rear-end and lane change conflict types. In the roundabout intersection used as a case study, there are
Section Rear-End Lane-Change no traffic signs that show the priority for the circulating traffic.
f % f % Moreover, the traffic approaching the intersection does not comply
with the priority rules unlike cases in developed countries. In addi-
1 199 9.5 % 104 32.4 %
2 261 12 % 3 0.9 %
tion, the entry and exit vehicle trajectories show that the drivers
3 461 22 % 200 62.3 % are crossing the circulating lanes to circulate in the intersection
4 792 38 % 13 4.0 % as close as possible to the circular island. This behavior results in
forming irregular congestion patterns in the intersection area as
shown in Fig. 9b, developing, traffic conflicts, as shown in Fig. 5.
maneuver performed by drivers in exit section 10 within a rela-
tively short distance (around 50 m). 4.3.2. Tailgating
For the roundabout intersection, driving behavior is studied Tailgating is the driving behavior where drivers are getting
using the captured vehicle trajectories for specific driving actions: as close as possible to the front vehicle [41]. This behavior
complying with priority rules, tailgating, and lane keeping. resulted in serious and sequenced rear-end conflicts in the
7
M. Shawky, A. Alsobky, A. Al Sobky et al. Ain Shams Engineering Journal 14 (2023) 102115

Fig. 7. Impact of centerline shifting.

Fig. 8. Impact of centerline shifting.

function area of the intersection from the same vehicle. that area of the intersection which includes about 82 % of the
can interpret the high number of rear-end conflict points com- recorded rear-end conflicts, as shown in Fig. 5 which illustrates
pared to the findings from developed countries. Drivers may the number and distribution of rear-end conflicts. Furthermore,
behave this way to avoid delays resulting from interference the vehicle trajectory shows that the number of rear-end con-
from surrounding vehicles who do not respect lane use priority flicts generated by a single vehicle in the intersection function
rules. However, this driving behavior is considered aggressive area is ranged from one to 13 conflicts with an average of
behavior; it is found to be a common behavior in the circular 2.7 rear-end conflicts/vehicle.

8
M. Shawky, A. Alsobky, A. Al Sobky et al. Ain Shams Engineering Journal 14 (2023) 102115

Fig. 9. Example of improper driving behavior.

4.3.3. Lane keeping Consequently, it is required to work on improving safety by pro-

In Egypt, drivers generally ignore lane-keeping behavior. They viding traffic safety education to the driver population and using
may use parts on two adjacent lanes while driving, especially on cutting-edge technologies in monitoring their behavior on differ-
curves, drivers may not stick to their lanes. This behavior may result ent road network elements. In addition, the findings show the
in dangerous safety situations where vehicles are getting close to importance of applying Road Safety Audits as a proactive safety
each other. Also, drivers may try to perform tailgating behavior to approach to avoid geometric safety problems. Furthermore, moni-
keep other vehicles out of their lanes. In addition, drivers could be toring the conflicts in a real-time and offline manner to keep track
pushed to use more than one lane while driving on curves as the lane of conflicts severity and provide feasible solutions.
width is not designed to accommodate different vehicle types such
as buses, trucks, etc. These types of vehicles require wider lane Declaration of Competing Interest
widths at curves which is not available in the current roundabout
intersection that has a lane width of around 3.5 m [43]. Therefore, The authors declare that they have no known competing finan-
this type of behavior may result in rear-end serious conflicts. cial interests or personal relationships that could have appeared
to influence the work reported in this paper.

5. Conclusion
References
Due to the limitation of the crash data regarding availability, [1] World Health Organization (WHO). Global status report on road safety 2018.
accuracy, and underreporting of crash events, applying a new sur- Available at https://www.who.int/publications/i/item/WHO-NMH-NVI-18.20.
rogate safety assessment process attracts more attention, espe- [2] World Health Organization (WHO). Fact sheet on Sustainable Development
Goals (SDGs): health targets. Available at https://apps.who.int/iris/bitstream/
cially in the era of big data that could be generated using drone handle/10665/340856/WHO-EURO-2017-2385-42140-58053-eng.pdf?
video analysis and connected automated vehicles. These data could sequence=1&isAllowed=y.
be used for safety assessment where traffic conflict and driving [3] Perkins SR, Harris JL. Criteria for traffic conflict characteristics. General Motors
Corporation; 1967. Report GMR 632.
behavior studies are applied. These techniques are promising [4] Fu C, Sayed T. Random parameters Bayesian hierarchal of traffic conflict
approaches for evaluating the current situation of traffic safety in extremes for crash estimation. Accid Anal Prev 2021;157:106159.
an off- and online manner. This paper employs an application of [5] Zheng L, Huang Y, Sayed T, Wen C. Validating the bayesian hierarchical
extreme value model for traffic conflict-based crash estimation on freeway
the traffic conflict technique to assess road safety in the round- segments with site-level factors. Accid Anal Prev 2021;159:106269.
about intersection and capture the road geometric deficiencies [6] Arun A, Haque MM, Bhaskar A, Washington S. Transferability of multivariate
and improper driver behavior in a developing country. A video extreme value models for safety assessment by applying artificial intelligence-
based video analytics. Accid Anal Prev 2022;170:106644.
was captured on a roundabout intersection in New Cairo City using
[7] Mannering F, Bhat C, Shankar V, Abdel-Aty M. Big data, traditional data, and
drone. The video processing technique is used to extract the vehi- the tradeoffs between prediction and causality in highway-safety analysis.
cle trajectories for sample vehicles in the recorded video. Then, the Anal Methods Acc Res 2020;25:100113.
vehicle trajectories are developed and analyzed to determine the [8] Xie K, Yang D, Ozbay K, Yang H. Use of real-world connected vehicle data in
identifying high-risk locations based on a new surrogate safety measure. Accid
conflict points in terms of location, PET, and angle. These data Anal Prev 2019;125:311–9.
are further analyzed to understand the effect of the interaction [9] Virdi N, Grzybowska H, Waller S, Dixit V. A safety assessment of mixed fleets
between the intersection geometry and driving behavior on traffic with Connected and Autonomous Vehicles using the Surrogate Safety
Assessment Module. Accid Anal Prev 2019;131:95–111.
safety in the intersection. [10] Naji J, Djebarni R. Shortcomings in road accident data in developing countries,
The main findings of this study could be divided into two main identification and correction: a case study. IATSS Res 2000;24(2).
categories which are the geometric design and the driver behavior. [11] Hu Y, Li Y, Huang H, Lee J, Youn J, Zou G. A high-resolution trajectory data-
driven method for real-time evaluation of traffc safety. Accid Anal Prev
The first category describes the effect of centerline shifting, lane 2022;165:106503.
balance, lane width, and entry and exit angles on intersection [12] Timmermans C, Shawky M, Alhajyaseen W, Nakamura H. Investigating the
safety. The second category describes the effect of the driver’s attitudes of Egyptian drivers toward traffic safety. IATSS Res 2022;46(1):73–81.
[13] Bayomi A, Shawky M, Okail M, Osama A. Investigating pedestrian safety-
behavior on the safety of the intersection as a response to safe driv- related behavior in developing countries: Egypt as a case study RSS(2022)
ing rules illiteracy in addition to deficiencies of the intersection special issue. Traffic Saf Res 2022;3. 10.55329/htwx5986.
geometry. The main improper driving behavior includes tailgating, [14] Islam Z, Adel-Aty M, Goswamy A, Abdelraouf A, Zheng O. Modelling the
relationship between post encroachment time and signal timings using UAV
lane overlapping, and improper compliance with priority rules. As
video data. arXiv:2210.05044, 2022.
a result, this study highlights the improper driver behavior that [15] Anarkooli AJ, Persaud B, Milligan C, Penner J, Saleem T. Incorporating speed in
could result from a lack of safe driving education and respond to a traffic conflict severity index to estimate left turn opposed crashes at
deficiencies in the geometric design. In general, these findings signalized intersections. Transp Res Rec 2021;2675(5):214–25.
[16] Peesapati LN, Hunter MP, Rodgers MO. Evaluation of post encroachment time
prove the interaction between intersection geometry, drivers’ as surrogate for opposing left-turn crashes. Transport Res Rec: J Transport Res
behavior, and road safety. Board 2013;2386:42–51.

9
M. Shawky, A. Alsobky, A. Al Sobky et al. Ain Shams Engineering Journal 14 (2023) 102115

[17] Zheng L, Sayed T, Mannering F. Modeling traffic conflicts for use in road safety [34] Zheng L, Tang M, Chen Y, Zhu G, Wang J, Lu H. Improving multiple object
analysis: a review of analytic methods and future directions. Anal Methods tracking with single object tracking. In IEEE/CVF conference on computer
Accid Res 2021;29:100142. vision and pattern recognition (CVPR); 2021.
[18] Wang C, Xie Y, Huang H, Liu P. A review of surrogate safety measures and their [35] Pallapotu K. data set generation using deep learning algorithms and visual
applications in connected and automated vehicles safety modeling. Accid Anal feature tracking. Master of Science Thesis, Michigan Technological University;
Prev 2021;157:106157. 2019.
[19] Arun A, Haque M, Washington S, Sayed T, Mannering F. A systematic review of [36] Lukezic A, Vojır T, Cehovin L, Matas J, Kristan M. Discriminative correlation
traffic conflict-based safety measures with a focus on application context. Anal filter with channel and spatial reliability. Int J Comput Vis 2018;126:671–88.
Methods Accid Res 2021;32:100185. [37] https://livecodestream.dev/post/object-tracking-with-opencv/#: :text=
[20] Essa M, Sayed T. Traffic conflict models to evaluate the safety of signalized Object%20tracking%20using%20OpenCV%20is,CSRT%2C%20GOTURN%2C%
intersections at the cycle level. Transp Res C 2018;89:289–302. 20and%20MediandFlow, visited on 11/10/2022.
[21] Essa M, Sayed T. Comparison between surrogate safety assessment model and [38] Kuipers T, Arya D, Gupta D. Hard occlusions in visual object tracking. In:
real-time safety models in predicting field-Mesures conflicts at signalized Bartoli A, Fusiello A, editors. Computer vision – ECCV 2020 workshops. ECCV
intersections. Transp Res Rec 2020;2674(3):100–12. 2020. Lecture notes in computer science. Vol. 12539. Springer, Cham; 2020.
[22] Yuan J, Abdel-Aty M. Approach-level real-time crash risk analysis for doi: 10.1007/978-3-030-68238-5_22.
signalized intersections. Accid Anal Prev 2018;119:274–89. [39] Taufique A, Minnehan B, Savakis A. Benchmarking deep trackers on aerial
[23] Fu C, Sayed T. Multivariate Bayesian hierarchical gaussian copula modeling of videos. Sensors 2020;20:547.
traffic conflict extremes for crash estimation. Anal Methods Accid Res [40] Pichaikuppan V, Narayanan R, Rangarajan A. Change detection in the presence
2021;29:100154. of motion blur and rolling shutter effect. In: Fleet D, Pajdla T, Schiele B,
[24] Fu C, Sayed T. A Multivariate method for evaluating from conflict extremes in Tuytelaars T, editors. Computer vision – ECCV 2014. ECCV 2014. Lecture notes
real time. Anal Methods Accid Res 2022;29:100244. in computer science, Vol. 8695. Springer, Cham; 2014.
[25] Abed HM, Ewadh HA. Coupling visual simulation model (VISSIM) with [41] Xiang H, Zhu J, Liang G, Shen Y. Prediction of dangerous driving behavior based
surrogate safety assessment model (SSAM) to evaluate safety at signalized on vehicle motion state and passenger feeling using cloud model and elman
intersections. J Phys Conf Ser 2021. neural network; 2021.
[26] Federal Highway Administration Research and Technology. Surrogate safety [42] Abojaradeh M, Jrew B, Al-Ababsah H, Al-Talafeeh A. The effect of driver
assessment model and validation: final report. Coordinating, Developing, and behavior mistakes on traffic safety. Civ Environ Res 2014;6(1):pp.
Delivering Highway Transportation Innovations, EEUU; 2016. [43] FHWA publications. Roundabouts: an informational guide. Report number
[27] De Ceunynck T. Defining and applying surrogate safety measures and FHWY-RD-00-067; 2000.
behavioural indicators through site-based observations. Hasselt, [44] A policy on geometric design of highways and streets. 5th edition.
Belgium: Hasselt University; 2017. Washington, DC: American Association of State Highway and Transportation
[28] Bonela SR, Kadali BR. Review of traffic safety evaluation at T-intersections Officials; 2004.
using surrogate safety measures in developing countries context”. IATSS Res
2022;46:307–21.
[31] Zheng O, Abdel-Aty M, Yue L, A, Wang Z, Mahmoud N. A drone-based vehicle
trajectory dataset for safety oriented research and digital twins. Comput Sci; Further reading
2022. Available: doi: 10.48550/arXiv.2208.11036.
[32] https://www.mathworks.com/discovery/object-detection.html visited on 11/ [29] Killi DV, Vedagiri P. Proactive evaluation of traffic safety at an unsignalized
10/2022. intersection using micro-simulation. J Traffic Logist Eng 2014;2(2):140–5.
[33] Jiménez-Bravo D, Murciego A, Mendes A, Blás H, Bajo J. Multi-object tracking ISSN 2315–4462.
in traffic environments: a systematic literature review. Neurocomputing [30] Paul M, Ghosh I. A Novel Approach of Safety Assessment at Uncontrolled
2022;494. Intersections using Proximal Safety Indicators. Eur Transp 2017;65(2). ISSN
1825–3997.

10