You are on page 1of 9

Transportation Letters

The International Journal of Transportation Research

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/ytrl20

Investigating dilemma zone boundaries for mixed


traffic conditions using support vector machines

Bharat Kumar Pathivada & Perumal Vedagiri

To cite this article: Bharat Kumar Pathivada & Perumal Vedagiri (2021): Investigating dilemma
zone boundaries for mixed traffic conditions using support vector machines, Transportation Letters,
DOI: 10.1080/19427867.2020.1870307

To link to this article: https://doi.org/10.1080/19427867.2020.1870307

Published online: 11 Jan 2021.

Submit your article to this journal

Article views: 58

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=ytrl20
TRANSPORTATION LETTERS
https://doi.org/10.1080/19427867.2020.1870307

Investigating dilemma zone boundaries for mixed traffic conditions using support
vector machines
Bharat Kumar Pathivada and Perumal Vedagiri
Department of Civil Engineering, Indian Institute of Technology Bombay, Mumbai, India

ABSTRACT KEYWORDS
Crashes take place at intersections even when signalized due to the presence of dilemma zone, where the Signalized intersection;
driver feels uncertain whether to cross or stop at the trigger of yellow signal. Erroneous decisions taken by yellow light; dilemma
the driver leads to rear-end or right-angle crashes, causing safety implications. Accurate position of the boundaries; support vector
machine (SVM); mixed traffic
dilemma zone can help in implementing various safety measures and enhance the safety at signalized
conditions
intersections. This study evaluates the dilemma boundaries for mixed traffic conditions using Support Vector
Machine (SVM), a supervised machine learning technique. The study results indicate that the position of the
dilemma zone is influenced by traffic and geometric characteristics. Results show that trucks and motorized
three-wheelers have dilemma zone closer to the intersection compared to other vehicle types. The dilemma
zone for mixed traffic conditions is found to be in the range of 30 m to 80 m from the stop line. The outcome
of this study contributes to better understand the dilemma boundaries in developing nations with mixed
traffic conditions, like India.

Introduction expressed in terms of distance to stop line (DSL) or time to stop line
(TSL) (Sharma 2008).
Traffic signals play a significant role in ensuring safe and efficient
movement of traffic at the intersections. Without signals, maneu­
vering would be exceedingly difficult in today’s congested roads Distance to stop line (DSL)
with the increasing vehicular population. The report ‘Road accident
Zegeer and Deen (1978) were the first to define to the dilemma
statistics in India 2018’ quotes that around 13,726 accidents took
boundaries using manually recorded driver responses at high-
place at the intersections with traffic light signal in India (MoRTH
speed-signalized intersections. According to Zegeer and Deen,
2019). Which is a big concern, as traffic signals are considered
‘the dilemma zone is defined as the approach area within which
efficient in controlling the traffic and reducing the conflicts at the
the probability of deciding to stop at the onset of yellow is within
intersections. Among possible factors contributing to the traffic
the range of 10 to 90%’. Li (2009) established the dilemma bound­
signal-related crashes, the presence of dilemma zone, upstream of
aries based on the 5% and 95% stopping probability of the drivers
the signalized approaches is one of the major causes. When vehicles
using a binary logistic regression model based on the observed
approach a signalized intersection at the onset of yellow signal, they
vehicle trajectories at four signalized intersections. Liu et al.
are often caught in an area where they must quickly decide whether
(2007) classified the driver behavior into three groups based on
to proceed or stop, which is termed as ‘dilemma zone’ in literature
their course of action and examined the variation of dilemma zone
(Urbanik and Koonce 2007; Wei, Li, and Ai 2009; Ni and Li 2014;
boundaries for the different driver groups. Based on results, authors
Zhang, Fu, and Hu 2014). An inappropriate decision might lead to
argued that there was significant discrepancy between the theoreti­
rear-end (stopping abruptly) or right-angle (failing to stop safely)
cally calculated dilemma zone and actual distribution of dilemma
collision (Hurwitz et al. 2012; Rahman et al. 2018), hampering the
zone at the signalized intersections.
safety at signalized intersections. Studies have stated that the
dilemma zone exists at the start of every yellow signal irrespective
of the duration of the yellow, as it depends on the complex decision- Time to stop line (TSL)
making behavior of the driver (Sheffi and Mahmassani 1981;
Urbanik and Koonce (2007) reviewed results from various studies
Urbanik and Koonce 2007). Considering the safety complications
on dilemma zone boundaries and suggested that the dilemma zone
due to the presence of the dilemma zone, it is vital to investigate the
tends to be about 2.5 to 5.5 seconds from the stop line. Since then,
position and boundaries of the dilemma zone at the signalized
zone limits of 2.5 to 5.5 seconds from the stop line, have been
intersection approaches.
popularly adopted in traffic safety domain (Hurwitz, Knodler, and
Nyquist 2011; Hurwitz et al. 2012; Zhang, Fu, and Hu 2014).
Literature review Hurwitz, Knodler, and Nyquist (2011) evaluated different speed
values and considered the 85th percentile speed as the input speed
Scanning the existing studies, there is scant research on computing
value to calculate the dilemma boundaries using the zone limits of
the dilemma zone boundaries at the signalized approaches. In
2.5 to 5.5 seconds from the stop line. Rakha, Amer, and El-
literature, it can be found that the dilemma boundaries are generally
Shawarby (2008) found that the dilemma zone for female drivers

CONTACT Bharat Kumar Pathivada pathivada.bharathkumar@gmail.com Department of Civil Engineering, Indian Institute of Technology Bombay, Mumbai
400076, India
© 2021 Informa UK Limited, trading as Taylor & Francis Group
2 B. K. PATHIVADA AND V. PERUMAL

is closer to the intersection compared to the male drivers. Similarly, To contend with these challenges, this study investigates the
older drivers had larger dilemma zone compared to younger dri­ dilemma zone boundaries using contemporary machine learning
vers. Considering the driver demographics Rakha, Amer, and El- technique.
Shawarby (2008) suggested that the dilemma zone zone should be
expanded and shifted from the current range of 2.5 to 5.5 seconds
range to 1.5 to 5.0 seconds from the stop line. Methodology
Data collection
Mixed traffic scenario To investigate the position of dilemma zone under mixed traffic
Although various researchers have modeled the dilemma driver conditions, five signalized intersection approaches were selected in
behavior considering vehicle types (Wei, Li, and Ai 2009; Gates Mumbai, India. The selected study locations were of typical 3-
and Noyce 2010; Pathivada and Perumal 2017). But not many legged and 4-legged intersections with similar traffic characteristics.
significant studies are found addressing the dilemma zone bound­ General characteristics of the selected study sections are shown in
aries at signalized intersections for mixed traffic conditions. Few Table 1. Traffic movement, driver actions in response to the yellow
studies can be found in literature considering dilemma zone for trigger were recorded throughout the study period by installing
mixed traffic scenario, but these studies have reported dilemma high definition cameras at vantage points (e.g. High-rise building,
boundaries for uncontrolled crossings and midblock pedestrian foot over bridge, etc.). Traffic cones were positioned at 10 m inter­
crossings (Pawar et al. 2016; Pawar and Patil 2017). Pawar and val, upstream of the stop line as reference points to obtain the
Patil (2017) addressed the dilemma of minor road vehicles intend­ approach speed and exact position of a vehicle, at the onset of
ing to cross the major road of un-signalized intersections based on solid yellow indication (shown in Figure A1, Appendix). Video
gap acceptance concept. data at all the selected approaches were collected during non-peak
hours (11am – 3pm), as it was observed that during peak hours,
dilemma behavior of the drivers could not be observed due to
Dilemma boundary evaluation techniques queuing.
Initially, a frequency-based approach was used to obtain the stop­
ping probability of the drivers (Zegeer and Deen 1978; Sharma Data processing
2008). The percentage of drivers stopping for a given distance (or
time) from the stop line and vehicle speed were used to develop the Video footage was studied frame by frame (at 30 frames per second)
dilemma zone curves. But studies showed significant variations in with the help of AVS video editor software to extract all the
the dilemma zone boundaries from the frequency-based approach necessary characteristics such as, precise timing of signal change,
(Sharma 2008). The variation in dilemma boundaries were attrib­ upstream position of the vehicle from stop line, type of vehicle and
uted to the complex driver decision behavior and also the frequency driver action (crossing/stopping at the stop line). Only through
approach was based on the weak law of large numbers (Sheffi and moving vehicles (first-to-go and last-to-stop at the trigger of yellow
Mahmassani 1981). So, later studies developed utility-based models signal) were considered in the extraction process. Turning vehicles
to explain the driver decision behavior at the trigger of yellow signal were not considered in the extraction as they tend to slow down at
(Köll, Bader, and Axhausen 2004; Papaioannou 2007; Kim et al. the intersection and do not exhibit dilemma behavior (Tarawneh
2008; Chiou and Chang 2010; Long, Han, and Yang 2011; Sharma, and Tarawneh 2002; Caird et al. 2007; ; Yang et al. 2014). Manual
Bullock, and Peeta 2011; Liu, Chang, and Yu 2012; Pathivada and extraction of the collected video footage resulted in 908 driver
Perumal 2019) and used the existing definitions (10–90% stopping responses (stop/go) at the onset of solid yellow indication.
probability or 2.5–5.5 seconds from stop line) to estimate the Descriptive statistics of the extracted data are shown in Table 2.
dilemma boundaries. These driver responses were analyzed using a supervised machine
learning algorithm, Support Vector Machine (SVM) to estimate the
position of the dilemma zone under mixed traffic conditions.
Problem statement
To abridge, though the aforementioned studies reported the
Support vector machines (SVM)
dilemma boundaries, still inconsistency remains in the boundary
definitions and their application. The current boundary definition Support Vector Machines (SVM) are supervised machine learning
of 10–90% stopping probability at the onset is based on empirical models, widely used in regression and classification problems
investigations from the manually collected field data using fre­ (Jiang, Missoum, and Chen 2014). SVM classifies the available
quency-based-approach. Investigating the driver responses using data by finding the optimal hyperplane that differentiates the data
the contemporary techniques and effect of various factors on the classes. Figure 1 shows dataset with two classes plotted in 2-dimen­
dilemma driver behavior can provide a better perspective of the sional space, which can be easily separated. For example, a line can
position of dilemma zone at signalized intersections. Also, the be traced to separate the two data classes. Such a line, which
existing boundary definitions might not be suitable to the current separates the data is called a hyperplane. A hyperplane is a general­
traffic scenario with varying vehicle characteristics and driver beha­ ization of a plane, as the dataset is in two-dimension, it is called a
vior as compared to the previous studies. Especially considering the line. The fact that SVM can be used for a dataset with p-dimensions,
mixed traffic scenario prevailing in the developing nations, where it is conveniently called a hyperplane. A form of the equation
vehicles vary in their physical and operating characteristics (Gowri defining the hyperplane, separating the classes can be expressed as
and Sivanandan 2015; Patnaik et al. 2020). The dilemma zone issue shown in Equation 1:
becomes much more complex for mixed traffic scenario because of
w:x þ b ¼ 0 (1)
variability in the operating speed of different vehicle types. Due to
the significantly diverse driver behavior and vehicle characteristics, Where w is a weight vector, x is input vector and b is bias. But there
results obtained from previous studies cannot be transferred might be many hyperplanes that can separate the data. The objec­
directly and might not be applicable to Indian traffic conditions. tive of SVM is to find the optimum hyperplane, which correctly
TRANSPORTATION LETTERS 3

Table 1. General characteristics of the selected study locations.


Volume
No. Intersection Location Layout No. of Approach Lanes (Vehicles/hour) Yellow Duration Cycle Length
1 Godrej Junction 3-Legged 5 4860 5 sec 130 sec
2 Tagore Junction 4-Legged 5 3665 3 sec 100 sec
3 Palm Road Junction 3-Legged 4 2832 4 sec 115 sec
4 CST Junction 4-Legged 3 1905 3 sec 120 sec
5 Shyam Nagar Junction 3-Legged 3 1875 3 sec 115 sec

Table 2. Descriptive statistics of the extracted data.


Speed statistics (Kmph)
Sample, N (%) 15th percentile speed 85th percentile speed
Vehicle Characteristics
Passenger Car 48.2 48.2 81.8
Motorized Two-wheeler 33.5 40.2 71.1
Motorized Three-wheeler 8.1 37.2 60.9
Truck 10.2 38.6 65.9
Intersection Characteristics
3-Legged Intersection 62.5 45 79.4
4-Legged Intersection 37.5 41.4 64.3

classifies the data and generalizes the error with the unseen data.
SVM finds this optimal separating hyperplane (OSH) based on the
margin (m). Where margin (m) is defined as the separation between
the hyperplane and the closest data point for a given weight w and a
bias b.
But most of the times available data is not easily separable. To
deal with the non-separable data, kernel trick is applied. Kernel
trick uses a mapping function to transform the non-separable data
to a higher dimension space, where the data can be linearly separ­
able (Pawar et al. 2015). Also, SVM uses a slack variable (�i ) which
measures the degree of misclassification between the classes (Cortes
and Vapnik 1995). Support vectors are the data points that are
difficult to classify and are close to the decision boundary (shown
in Figure 1). Optimum hyperplane (Ho) is defined as a hyperplane
with a maximum margin of separation (m), obtained by minimizing
Figure 1. Dataset with optimal separating hyperplane.
the error function. Optimum separating hyperplane can be
obtained by optimizing Equation 2:
( )
jjwjj2 Xn
min : þC �i stop at the intersection, indicating two data classes. The two classes are
w;b;� 2 i¼0 (2) indicated as follows: positive (+1) for vehicles crossing the intersection
and negative (−1) for vehicles stopping at the intersection. To classify
Subject to yi ðw:xi þ bÞ � 1 �i for i ¼ 1 . . . ; N
the stop/go responses, driver actions based on their position with
Where C is the cost coefficient, and w is the weight of each feature. respect to stop line and approach speed of the subject vehicle are
k-fold cross-validation technique is commonly used to obtain the plotted. The two classes (stop/cross) in our dataset are non-separable
parameters of an SVM model (Jiang, Missoum, and Chen 2014). In and the optimal separating hyperplane is constructed by the SVM
this technique, data from both the classes are randomly divided into algorithm. Dilemma zone is an area, where most of the driver responses
k subsets of equal size. Where, k – 1 subsets are used to build the are overlapping i.e., where the driver has an option to cross or stop at
model, and the remaining subset is used to validate the developed the intersection.
model. The process is then repeated k times, with each subset used Once the separating hyperplane for the profiles is found, the
once as a validation sample. Results from the k-folds are averaged to position of the dilemma zone is obtained by considering the driving
obtain the model parameters. population within 15th and 85th percentile speeds. As considering
100% driving population might over-estimate the boundaries of
dilemma zone, and it is generally assumed that the upper and
Anaysis and results lower 15 percentile speeds represent vehicles that are either too
fast or too slow for existing conditions. The 85th and 15th percentile
Analysis of dilemma boundaries using SVM
speeds are commonly used statistical metrics and give an overall
Dataset is composed of N sets of elements ðxi ; yi Þ, where xi , is description of the high and low speeds observed by the most
a p-dimensional vector and each xi is associated with a value yi reasonable drivers.
indicating, if the element belongs to the class (+1) or not (−1). (yi can This study estimates the upper and lower dilemma boundaries
only have two possible values −1 or +1, indicating the class). In our based on the 85th percentile and 15th percentile speeds observed at
dataset, when a driver faces a yellow signal, he/she must either cross or the intersection approaches. The considered study locations have
4 B. K. PATHIVADA AND V. PERUMAL

mixed traffic scenario typically seen in developing nations, where a Driver responses for all four vehicle categories, optimum
wide variety of vehicles with varying physical and dynamic char­ separating hyperplane obtained from SVM algorithm are
acteristics, ply on the roads for varying needs. shown in Figure 2. The results establish that the trucks and
motorized three-wheelers (MTW) have dilemma zone closer to
the intersection (i.e., they delay their stop-go decision). This is
Dilemma zone boundaries for vehicle type likely because trucks and MTWs have lower operating speeds
compared to other vehicle types. Passenger cars and motorized
To understand the distribution of dilemma zone boundaries across
two-wheelers (M2W) were found to have dilemma zone far
vehicle types, driver responses from different vehicle types (passen­
away from the stop line compared to trucks and MTWs. This
ger car, motorized two-wheeler, motorized three-wheeler, and
might be because they generally have higher operating speeds
truck) were analyzed using the SVM algorithm. Figure 2 shows
and are aggressive in nature. Boundaries and length of the
the driver responses for different vehicle categories, green and red
dilemma zone for all the vehicle types from the analysis are
profiles shown are the cross and stop responses, respectively. The
shown in Table 3.
optimal separating hyperplane (OSH) is found for the overlapping
profiles by the SVM algorithm, and the circular profiles identified
by the algorithm are support vectors. The boundaries of the Dilemma zone boundaries for intersection type
dilemma zone are obtained in terms of distance to stop line (DSL)
by projecting lines corresponding to the 15th percentile and 85th Earlier studies (Pathivada and Perumal 2017, 2019) found that
percentile speeds observed (from Table 2) at the intersection ‘type of intersection’ has a significant impact on the driver’s
approaches. Distance from stop line corresponding to the 15th decision behavior at the yellow trigger. This study further ana­
percentile speed is considered as lower boundary (LB) of the lyzed the impact of intersection type (3-legged and 4-legged) on
dilemma zone and the distance corresponding to the 85th percentile the position of the dilemma zone. Optimum separating hyper­
speed is considered as the upper boundary (UB) of the dilemma planes for the driver responses at 3-legged and 4-legged
zone. Length of dilemma zone is the difference between the lower approaches were found using the SVM algorithm (shown in
and upper boundaries. Figure 3). Results establish that 3-legged approaches had larger
Considering the mixed traffic conditions, there was significant dilemma zone compared to the 4-arm approaches. A possible
variation in the operating speeds of different vehicle types. So, the reason could be that drivers are willing to take higher risks at 3-
prevailing 15th and 85th percentile speeds of individual vehicle types legged intersections because of a fewer number of conflicts
were considered for evaluating the lower and upper boundaries of involved. The dilemma boundaries for 3-legged intersection
dilemma zone. were found to be as 47 m and 86 m, lower and upper bound­
aries, respectively. Whereas the dilemma boundaries for 4-arm

Lower Boundary Upper Boundary

(a) (b)

Optimum
Hyperplane

(c) (d)

Figure 2. Dilemma boundaries for different vehicle types (a) Passenger Car, (b) Motorized two-wheeler, (c) Motorized three-wheeler, and (d) Truck.
TRANSPORTATION LETTERS 5

Table 3. Distribution of dilemma boundaries under mixed traffic conditions. importance to implement various countermeasures to assist drivers
Passenger Car M2W MTW Truck in their decision making. Although dilemma zone cannot be com­
Lower Dilemma Boundary (LB) 45 m 45 m 30 m 33 m pletely eliminated, (as it depends on the driver decision behavior)
Upper Dilemma Boundary (UB) 80 m 75 m 67 m 76 m safety complications can be reduced at the signalized intersections
Length of Dilemma Zone 35 m 30 m 37 m 43 m by providing safety measures such as additional pavement marking,
M2W: Motorized two-wheeler; MTW: Motorized three-wheeler advance warning signs, etc.

intersection were found to be 40 m and 50 m, lower and upper Summary and conclusions
boundaries, respectively.
This study investigated the position and distribution of dilemma
zone for mixed traffic conditions. Although, previous studies have
Discussion attempted to quantify the dilemma boundaries, due to the signifi­
cantly diverse behavior of Indian drivers implementing the results
The study results show that the traffic characteristics and roadway directly from the previous studies might not suffice. To compute the
geometrics have a significant impact on the position of dilemma dilemma boundaries, driver actions in response to the signal change
zone. The distribution of the dilemma zone boundaries for different were video tapped at five signalized intersection approaches. This
vehicle types is shown in Table 3. The variations in the boundaries study employed a machine learning algorithm (Support Vector
of the dilemma zone is likely due to their variations in the operating Machine) to classify the driver responses at the yellow onset. SVM
speeds of the vehicles and their manoeuvrability. In the mixed algorithm generates an optimum separating hyperplane (OSH)
traffic scenario, different vehicles co-exist on the same carriageway, which separates the two data classes (stop/cross) of driver
so these variations must be taken into consideration while design­ responses. Dilemma zone, which is an area ahead of the intersec­
ing dilemma protection systems. In this regard, the design of safety tion, where the driver has an option to stop/go at the trigger of
systems for the dilemma zone under mixed traffic conditions can be yellow signal is defined based on the observed 85th and 15th per­
considered in the range of 30 m to 80 m from the stop line. Location centile speeds. The boundaries of the dilemma zone in this study
of the dilemma zone at an intersection approach is depicted in were defined in terms of distance to the stop line (DSL). The upper
Figure 4. The precise location of the dilemma zone is of great boundary (UB) and lower boundary (LB) of the dilemma zone were

(a) (b)

Figure 3. Dilemma boundaries for (a) 3-legged, and (b) 4-legged intersections.

Proposed dilemma zone (DZ) road marking

Figure 4. Location of dilemma zone at a signalized intersection approach.


6 B. K. PATHIVADA AND V. PERUMAL

then defined based on the 85th percentile and 15th percentile speeds Kim, W., J. Zhang, A. Fujiwara, T. Y. Jang, and M. Namgung. 2008. “Analysis of
observed at the intersection approaches. The driving population Stopping Behavior at Urban Signalized Intersections: Empirical Study in
South Korea.” Transportation Research Record 2080 (1): 84–91.
between the 15th and 85th percentile speeds are used to define the
doi:10.3141/2080-10.
dilemma zone since the lower and upper 15th percentile drivers are Köll, H., M. Bader, and K. W. Axhausen. 2004. “Driver Behaviour during
either too fast or too slow for the existing conditions. Flashing Green before Amber: A Comparative Study.” Accident Analysis &
Study results indicate that the location of the dilemma zone varies Prevention 36 (2): 273–280. doi:10.1016/S0001-4575(03)00005-8.
with the type of vehicle. It is logical, as the vehicles vary in their Li, Z. 2009. “Modeling Dynamic Dilemma Zones Using Observed Yellow-onset
Trajectories.” Institute of Transportation Engineers. ITE Journal 79 (11): 24.
physical and dynamic characteristics, which influences their man­ Liu, Y., G. L. Chang, R. Tao, T. Hicks, and E. Tabacek. 2007. “Empirical
oeuvrability. 3-legged approaches were found to have larger dilemma Observations of Dynamic Dilemma Zones at Signalized Intersections.”
zone compared to the 4-legged approaches, indicating drivers willing Transportation Research Record 2035 (1): 122–133. doi:10.3141/2035-14.
to take higher risk to cross at the 3-legged intersections. Dilemma Liu, Y., G. L. Chang, and J. Yu. 2012. “Empirical Study of Driver Responses
during the Yellow Signal Phase at Six Maryland Intersections.” Journal of
zone for the truck and motorized three-wheeler was found to be
Transportation Engineering 138 (1): 31–42. doi:10.1061/(ASCE)TE.1943-
closer to the stop line compared to other vehicle types. Passenger 5436.0000278.
car and motorized two-wheeler had dilemma zone away from the Long, K., L. D. Han, and Q. Yang. 2011. “Effects of Countdown Timers on
stop line, indicating their aggressive and risk-taking behavior. Based Driver Behavior after the Yellow Onset at Chinese Intersections.” Traffic
on the study results, dilemma zone for mixed traffic conditions tends Injury Prevention 12 (5): 538–544. doi:10.1080/15389588.2011.593010.
MoRTH, Ministry of Road Transport and Highways. 2019. “Road Accidents in
to be between 30 m to 80 m from the stop line. The accurate position India 2018.” www.morth.nic.in
of dilemma zone boundaries can help in developing various passive Ni, Y., and K. Li. 2014. “Estimating Rear-end Accident Probabilities at Signalized
and active dilemma protection systems (such as additional road Intersections: A Comparison Study of Intersections with and without Green
markings, road signs, advance warning systems etc.) to assist the Signal Countdown Devices.” Traffic Injury Prevention 15 (6): 583–590.
doi:10.1080/15389588.2013.845752.
drivers in their decision (stop/go) process and reducing the safety
Papaioannou, P. 2007. “Driver Behaviour, Dilemma Zone and Safety Effects at
complications. Results from this study contribute to better under­ Urban Signalised Intersections in Greece.” Accident Analysis & Prevention 39
stand the dilemma boundaries in developing countries with mixed (1): 147–158. doi:10.1016/j.aap.2006.06.014.
traffic conditions, like India. Although results obtained from this Pathivada, B. K., and V. Perumal. 2017. “Modeling Driver Behavior in Dilemma
study are only applicable to mixed traffic conditions, the defined Zone under Mixed Traffic Conditions.” Transportation Research Procedia 27:
961–968. doi:10.1016/j.trpro.2017.12.120.
methodology can be applied for a different dataset to obtain the Pathivada, B. K., and V. Perumal. 2019. “Analyzing Dilemma Driver Behavior at
desired results. Also, this study did not consider the effect of driver Signalized Intersection under Mixed Traffic Conditions.” Transportation
attributes such as age and gender, on the location of dilemma zone, as Research Part F: Traffic Psychology and Behaviour 60: 111–120. doi:10.1016/
these details could not be extracted from the video data. Further J.TRF.2018.10.010.
Patnaik, A. K., L. A. Agarwal, M. Panda, and P. K. Bhuyan. 2020. “Entry
research can be carried out to study the influence of turning move­
Capacity Modelling of Signalized Roundabouts under Heterogeneous
ments on the driver behavior and the effect of driver attributes on the Traffic Conditions.” Transportation Letters 12 (2): 100–112. doi:10.1080/
location of dilemma zone. Also, further studies are required to 19427867.2018.1533160.
evaluate the dilemma boundaries using different percentile speeds Pawar, D. S., V. Kumar, N. Singh, and G. R. Patil. 2016. “Analysis of Dilemma
and validate the dilemma zone boundaries. Zone for Pedestrians at High-speed Uncontrolled Midblock Crossing.”
Transportation Research Part C: Emerging Technologies 70: 42–52.
doi:10.1016/j.trc.2016.04.012.
Disclosure statement Pawar, D. S., and G. R. Patil. 2017. “Minor-Street Vehicle Dilemma while
Maneuvering at Unsignalized Intersections.” Journal of Transportation
No potential conflict of interest was reported by the authors. Engineering, Part A: Systems 143 (8): 04017039. doi:10.1061/
JTEPBS.0000066.
Pawar, D. S., G. R. Patil, A. Chandrasekharan, and S. Upadhyaya. 2015.
References “Classification of Gaps at Uncontrolled Intersections and Midblock
Crossings Using Support Vector Machines.” Transportation Research
Caird, J. K., S. L. Chisholm, C. J. Edwards, and J. I. Creaser. 2007. “The Effect of Record 2515 (1): 26–33. doi:10.3141/2515-04.
Yellow Light Onset Time on Older and Younger Drivers’ Perception Response Rahman, Z., D. Martinez, N. Martinez, Z. Zhang, A. Memarian, S. Pulipati, S. P.
Time (PRT) and Intersection Behavior.” Transportation Research Part F: Mattingly, and J. M. Rosenberger. 2018. “Evaluation of Cell Phone Induced
Traffic Psychology and Behaviour 10 (5): 383–396. doi:10.1016/j.trf.2007.03.002. Driver Behavior at a Type II Dilemma Zone.” Cogent Engineering 5 (1):
Chiou, Y. C., and C. H. Chang. 2010. “Driver Responses to Green and Red 1436927. doi:10.1080/23311916.2018.1436927.
Vehicular Signal Countdown Displays: Safety and Efficiency Aspects.” Rakha, H., A. Amer, and I. El-Shawarby. 2008. “Modeling Driver Behavior
Accident Analysis & Prevention 42 (4): 1057–1065. doi:10.1016/j. within a Signalized Intersection Approach Decision–Dilemma Zone.”
aap.2009.12.013. Transportation Research Record 2069 (1): 16–25. doi:10.3141/2069-03.
Cortes, C., and V. Vapnik. 1995. “Support-vector Networks.” Machine Learning Sharma, A. 2008. “Integrated behavioral and economic framework for improv­
20 (3): 273–297. ing dilemma zone protection systems.” Doctoral dissertation, Purdue
Gates, T. J., and D. A. Noyce. 2010. “Dilemma Zone Driver Behavior as a University.
Function of Vehicle Type, Time of Day, and Platooning.” Transportation Sharma, A., D. Bullock, and S. Peeta. 2011. “Estimating Dilemma Zone Hazard
Research Record 2149 (1): 84–93. doi:10.3141/2149-10. Function at High Speed Isolated Intersection.” Transportation Research Part
Gowri, A., and R. Sivanandan. 2015. “Evaluation of Right-turn Lanes at C: Emerging Technologies 19 (3): 400–412. doi:10.1016/j.trc.2010.05.002.
Signalized Intersection in Non-lane-based Heterogeneous Traffic Using Sheffi, Y., and H. Mahmassani. 1981. “A Model of Driver Behavior at High Speed
Microscopic Simulation Model.” Transportation Letters 7 (2): 61–72. Signalized Intersections.” Transportation Science 15 (1): 50–61. doi:10.1287/
doi:10.1179/1942787514Y.0000000034. trsc.15.1.50.
Hurwitz, D. S., M. A. Knodler Jr, and B. Nyquist. 2011. “Evaluation of Driver Tarawneh, T. M., and M. S. Tarawneh. 2002. “Compliance and Comprehension
Behavior in Type II Dilemma Zones at High-speed Signalized Intersections.” of the Yellow Signal Indication: A Case Study from Jordan.” Traffic Injury
Journal of Transportation Engineering 137 (4): 277–286. doi:10.1061/(ASCE) Prevention 3 (4): 298–302. doi:10.1080/15389580214627.
TE.1943-5436.0000219. Urbanik, T., and P. Koonce. 2007. “The Dilemma with Dilemma Zones.”
Hurwitz, D. S., H. Wang, M. A. Knodler Jr, D. Ni, and D. Moore. 2012. “Fuzzy Proceedings of ITE District, 6th Annual Meeting, Portland, Oregon.
Sets to Describe Driver Behavior in the Dilemma Zone of High-speed Wei, H., Z. Li, and Q. Ai. 2009. “Observation-based Study of Intersection
Signalized Intersections.” Transportation Research Part F: Traffic Psychology Dilemma Zone Natures.” Journal of Transportation Safety & Security 1 (4):
and Behaviour 15 (2): 132–143. doi:10.1016/j.trf.2011.11.003. 282–295. doi:10.1080/19439960903441141.
Jiang, P., S. Missoum, and Z. Chen. 2014. “Optimal SVM Parameter Selection for Yang, Z., X. Tian, W. Wang, X. Zhou, and H. Liang. 2014. “Research on Driver
Non-separable and Unbalanced Datasets.” Structural and Multidisciplinary Behavior in Yellow Interval at Signalized Intersections.” Mathematical
Optimization 50 (4): 523–535. doi:10.1007/s00158-014-1105-z. Problems in Engineering 2014. doi:10.1155/2014/518782.
TRANSPORTATION LETTERS 7

Zegeer, C. V., and R. C. Deen. 1978. “Green-Extension Systems At High-Speed Zhang, Y., C. Fu, and L. Hu. 2014. “Yellow Light Dilemma Zone Researches: A
Intersections.” Kentucky Department of Transportation (Research Report Review.” Journal of Traffic and Transportation Engineering (English Edition)
496). 1 (5): 338–352. doi:10.1016/S2095-7564(15)30280-4.
8 B. K. PATHIVADA AND V. PERUMAL

Appendix

Figure A1. Field data collection setup (Pathivada and Perumal 2017).

You might also like