International Journal of Transportation Science and Technology 7 (2018) 137–150

Contents lists available at ScienceDirect

International Journal of Transportation

Science and Technology
journal homepage: www.elsevier.com/locate/ijtst

Automated intersection delay estimation using the input–

output principle and turning movement data
Ibrahem Shatnawi a, Ping Yi b,⇑, Ibrahim Khliefat c
a
CH2M Hill, Two Easton Oval, Suite 500, Columbus, OH 43219, United States
b
Department of Civil Engineering, The University of Akron, Akron, OH 44325-3905, United States
c
Department of Civil Engineering, Al-Balqa Applied University, Jordan

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

Article history: Vehicle delay is one of the most important performance measures of effectiveness (PMOE)
Received 24 October 2017 in intersection traffic operations. It allows traffic engineers to evaluate the performance of
Received in revised form 28 March 2018 a traffic system component or the effectiveness of the system-wide control strategy. It is
Accepted 7 April 2018
often utilized for real-time applications such as adaptive signal control, congestion man-
Available online 13 April 2018
agement, and dynamic traffic assignment. However, obtaining intersection performance
data in real time such as intersection control delay can be very time consuming and labor
Keywords:
intensive. This paper develops and tests a reliable method called the Automated Vehicle
Automated delay estimation
Detection system
Delay Estimation Technique (AVDET), which automatically estimates delay at a signalized
Intersection delay intersection using detector data and signal timing information from the existing traffic sig-
Approach delay nal controller. Results from the delay estimation algorithm were compared with those from
simulation, followed by statistical tests under varying traffic operation conditions. The
findings showed that AVDET was able to provide effective results under different traffic
and signal control scenarios. Future work of field implementation for the proposed algo-
rithms is recommended to investigate the model reliability and effectiveness in real-
time traffic conditions.
Ó 2018 Tongji University and Tongji University Press. Publishing Services by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).

1. Introduction

Vehicle delay is considered as one of the most important performance measures of effectiveness (PMOE) in intersection
traffic operations because it allows traffic engineers to evaluate the overall performance of a traffic system. Currently, vehicle
delay is used as a principal performance measure to determine intersection level of service (LOS), estimate average speed,
and calculate fuel consumption and emissions. Therefore, it is essential to develop a reliable method to accurately measure
delay in real time.
According to the Transportation Research Board’s Highway Capacity Manual (HCM2010), control delay is defined as the
additional travel time experienced by a vehicle affected by intersection control (Highway Capacity Manual, 2010). As shown
in Fig. 1, control delay can be separated into different parts such as deceleration delay, stop delay, acceleration delay,
approach delay and intersection delay. The definitions for these terms are provided as follows:

Peer review under responsibility of Tongji University and Tongji University Press.
⇑ Corresponding author.
E-mail addresses: [email protected] (I. Shatnawi), [email protected] (P. Yi), [email protected] (I. Khliefat).

https://doi.org/10.1016/j.ijtst.2018.04.001
2046-0430/Ó 2018 Tongji University and Tongji University Press. Publishing Services by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
138 I. Shatnawi et al. / International Journal of Transportation Science and Technology 7 (2018) 137–150

Fig. 1. Delays at signalized intersection.

Control delay is the total delay experienced by a vehicle due to intersection control.

Approach delay is the delay experienced by a vehicle before entering the intersection.

Intersection delay is the additional travel time experienced by a vehicle after it enters the intersection and before it reaches
free-flow speed.

Deceleration delay is defined as the delay experienced by a vehicle when it is reducing its speed.

Stop delay is the waiting time for a vehicle when its speed is zero (practically, we consider a vehicle as stopped when its
speed is less than 5 mph) (William Walker, 1955).

Acceleration delay is defined as additional travel time experienced by a vehicle during the acceleration period.

Thus, control delay may be obtained as:

control delay ¼ deceleration delay þ stop delay þ acceleration delay ð1Þ

To simplify,

control delay ¼ approach delay þ intersection delay ð2Þ

Delay can be estimated through field measurement, simulation, analytical derivation, or by using a combination of these
methods. Of these, simulation and analytical estimation models are the most convenient and are more commonly used, since
field experiments are costly and time-consuming to conduct. From a technical perspective, it is also very challenging to
obtain field measurements of approach delay, which consists of delay experienced by vehicles decelerating towards a queue
or red light, delays experienced by vehicles stopped at the intersection or in a queue, and delays experienced by departing
vehicles as they accelerate through the intersection and depart.
Although the simulation method is more convenient, less expensive and more flexible, the biggest challenge is to ensure
that the simulation platform is built to represent the field environment in a realistic manner and that the simulation model is
fully calibrated to replicate the individual behavior as well as the interactions of the vehicles at the intersection such that the
artificial process is not biased. Likewise, the application of analytical models to this type of engineering problem is often lim-
ited in effectiveness due to assumptions and constraints on the initial boundary conditions or oversimplifications to find fea-
sible and meaningful mathematical solutions.
In this study, we present a new approach to estimating vehicle delay at a signalized intersection. This method makes use
of the input–output principle as well as vehicle origin–destination (O–D) information obtained from the Automatic Turning
Movement Identification System (ATMIS) (Xu and Yi, 2013). Using this framework, approach delays can be estimated by ana-
lyzing real-time data obtained from the arrival and departure detectors located at the areas upstream and downstream from
an intersection. Also, intersection delay is calculated by measuring the time spent within the intersection by comparing the
time it takes a vehicle to pass between the paired detectors defining the origin and destination of each turning movement.
Laboratory simulation using the proposed method demonstrates that it is feasible, effective, and reliable in dealing with dif-
ferent traffic conditions (low, medium, high, and saturated), oversaturated conditions which may cause street gridlock have
not considered in this study. This paper presents the fundamentals of the proposed model, its modeling details and
implementation.
 I. Shatnawi et al. / International Journal of Transportation Science and Technology 7 (2018) 137–150 139

2. Literature review

The latest advances in detection and communication technologies have greatly enhanced the ability to obtain traffic
information in real time. This information supports intelligent transportation system (ITS) applications in traffic signal con-
trol and allows traffic engineers in traffic management centers (TMCs) to proactively respond to changes in traffic conditions
to prevent traffic congestion and increase throughput.
Various techniques have been proposed in the past for delay estimation at signalized intersections. Early methods sug-
gested by the Bureau of Public Roads’ Committee on Operating Speeds in Urban Areas in 1955 (William Walker, 1955)
involved continuously taking photographs of intersection approaches and calculating vehicle delays manually. More
recently, (Kinzel, 1992) suggested a method in the Traffic Engineering Handbook to calculate vehicle delay in the field by sam-
pling vehicles in a queue at a signalized intersection. This method, which is employed in HCM2010, relies on manual mea-
sures of stopped vehicles, slow-moving vehicles, and vehicles passing through the intersection at small time intervals; it uses
stop delay and adjustment factors to estimate the control delay. This method is very labor intensive, and its accuracy relies
highly on the judgment of the user regarding the status of each vehicle and the selection of adjustment factors.
Researchers (Quiroga and Bullock, 1999) attempted to use Global Positioning System (GPS) coordinates to estimate con-
trol delay. Their method records the speed and location of sampled vehicles every second; however, the small sample size (as
few vehicles had GPS devices) limited the collection of sufficient data. (Mousa, 2002) suggested a method for measuring and
analyzing control delay by tracking vehicles’ arrival time at checkpoints manually. In a field experiment, twelve screen lines
(checkpoints) were marked for one approach with a 27- to a 55-m gap between lines. Crossing times for all vehicles through
each screen were manually recorded using audio-cassette recorders. Although it is easy to implement in the field, it is dif-
ficult to apply this method for vehicle tracking in multilane situations where the volume of traffic is high.
A study (Abdel-Rahim and Dixon, 2009) produced an automated measurement of approach delay at signalized intersec-
tions. Delay estimation for all four approaches and lane groups at an intersection was carried out using video detection. This
method only allows the calculation of approach delay, and it was reported to be able to provide more accurate and less
biased delay estimations than those of HCM2010. Suggested a method (Teply, 1989) to measure approach delay in the field
using three timestamps: the arrival time of vehicles to a point located upstream of the approach, beyond the point that a
queue usually could be reached; the timestamp when the vehicle crosses the stop bar; and the timestamp at the beginning
or end of the green phase of the signal. This method is more amenable to accurate field implementation than sampling tech-
niques that require more complex data collection and processing. However, it cannot measure intersection delays, since the
turning movements of vehicles are not tracked as they enter and leave the intersection. Another similar study (Kebab and
Dixon, 2007) recommended installing additional detectors to cover all lanes separately. In this way, the approach delay
for different lane groups can be obtained. As the study did not identify vehicle turning movements, this technique cannot
be used to measure intersection control delay. In a recent study (Forbush, 2011) developed an automated method to esti-
mate traffic delay at an intersection in real time. However, the delay estimation focused on through lanes only, and situations
where lanes are shared were not considered in this study.
In summary, most existing approaches rely heavily on manual work or are not able to accurately measure intersection
control delay for traffic control applications. Driven by the need to overcome such limitations and to provide a further con-
tribution to this important area of research, we developed a method to automatically calculate both the approach delay and
intersection delay at a signalized intersection, which we call the Automated Vehicle Delay Estimation Technique (AVDET).
The following sections discuss the research methodology as well as the details of testing and data analysis to evaluate this
new technique.

3. Research methodology

The proposed methodology for estimating performance measures at isolated intersections characterizes vehicle delay at a
signalized intersection as a function of intersection delay (the time lost when vehicles cross the intersection) and approach
delay (the time lost during the process of vehicle deceleration and re-acceleration). The AVDET system can determine these
two performance measures. The following sub-sections provide details regarding how each component of vehicle delay is
defined and measured.

3.1. Approach delay

To demonstrate the dynamics of approach delay, it is assumed that an arrival detector is placed upstream of the intersec-
tion to avoid vehicle queue spillback, as shown in Fig. 2 where dashed green lines are used to represent the vehicle trajectory.
Two conditions influence the shape of the vehicle trajectory lines. In the first condition, a vehicle approaches the intersection
during the red phase of the traffic signal and/or when a queue exists, so the vehicle will slow down and start to accrue delay
due to deceleration as shown by the trajectory lines a and d. In the second condition, the vehicle arrives during the green
phase of the traffic signal, and the queue size is zero as shown in trajectory b; the vehicle will cross the intersection with
no approach delay. Nevertheless, vehicles entering the intersection during the green phase may further encounter intersec-
140 I. Shatnawi et al. / International Journal of Transportation Science and Technology 7 (2018) 137–150

Fig. 2. Vehicle delay scheme.

tion delay due to blocking by cross traffic or having to reduce speed in order to make a turn, as illustrated in trajectories c and
d, respectively.
The state of movement of a vehicle at a signalized intersection may be represented by matrix M ¼ fmj;k;Ui g, which is
defined by the following terms:

1. Ui is the phase number,

2. j is the direction of travel (NB, SB, EB, and WB), and
3. k is the turning lane group.

Each element of the traffic release matrix M can be described by:

0; the vehicle atðj; k; Ui Þ is occupying the detector during the red phase
mj;k;Ui ¼ ð3Þ
1; the vehicle atðj; k; Ui Þ is deactivating the detector due to departure
where Ui is equal to 1, 2, 3, or 4 (in the case of a four-phase operation) to represent different phases; j takes the value of 1, 2,
3, or 4 to represent traffic in the westbound, northbound, eastbound, and southbound directions, respectively. The element k
takes the value of 1, 2 or 3 to represent left, through, or right movement, respectively. Fig. 3 shows the state of movement for
a particular vehicle at a signalized intersection.
Following the above definitions, the traffic movement matrix M as depicted in Fig. 3 can be represented by the following
expression:
2 3
1 j1 k1 j1 k2 j1 k 3 j2 k1 j2 k 2 j2 k 3 j3 k1 j3 k 2 j3 k3 j4 k1 j4 k 2 j4 k3
6U 1 0 0 0 0 1 1 0 0 0 0 1 7
6 1 7
6 7
6 U2 0 1 1 0 0 0 0 1 1 0 0 0 7
mj;k ; Ui ¼ 6
6U
7 ð4Þ
6 3 1 1 1 0 0 0 0 1 1 0 0 0 7
7
6 7
4 U4 0 0 1 1 0 0 0 0 1 1 0 0 5
0 0 0 0 1 1 0 0 0 0 1 1
As vehicles approach an intersection, the arrival detectors and departure detectors (shown in Fig. 4) record the changes in
the queue size at each intersection approach due to signal control. The arrival detectors are placed a sufficient distance
upstream of the intersection so that it is not influenced by queue spillbacks.
Following the input–output principle, Eq. (5) can be used to determine approach delay by considering each signal phase:
q a
DUi;t ¼ dUi;t þ dUi;t  rUi;t ð5Þ

where
DUi ;t = approach delay for phase Ui in the ðtÞth second (veh-s)
 I. Shatnawi et al. / International Journal of Transportation Science and Technology 7 (2018) 137–150 141

Fig. 3. Event of vehicle flow at a signalized intersection: (a) geometry, (b) phase diagram, and (c) turning movement diagram.

Fig. 4. Detected vehicles at intersection approach.

142 I. Shatnawi et al. / International Journal of Transportation Science and Technology 7 (2018) 137–150

q
dUi ;t = delay due to queued vehicles in the ðtÞth second (veh-s)
a
dUi ;t = delay to newly arrived vehicles in the ðtÞth second (veh-s)
rUi ;t = delay reduction due to vehicle releases in the ðtÞth second (veh-s)

In queue delay calculations, the total number of vehicle arrivals and departures are tracked, second by second, using real-
time detection data (as described later). The delay due to the number of vehicles in the queue is equal to the product of the
existing queue size and the wait time. The queue delay must also consider the delay that has already been incurred in the
preceding time interval due to residual queue. Thus, Eq. (6) is formulated:
q
XX
dUi;t ¼ j k
lUi;t ð6Þ

where
j = approach number (1, 2, 3, or 4);
k = lane group of approach j (1, 2, or 3); and
lUi;t = queue size at end of ðUi;t Þth time interval (1 s) in lane group k of approach j(veh).

The delay resulting from new vehicle arrivals is found for each lane group, and the total arrival delay is calculated using
Eq. (7),
a
XX
dUi;t ¼ j k
kUi;t ð7Þ

where
kUi;t = arrival rate of vehicles in lane k of approach j
in the ðtÞth second of phase Ui (veh/s).

The reduction in approach delay due to vehicle departures is the product of the vehicle departure rate as shown by Eq. (8).
If the current phase of the traffic signal is not green during the approach, then its release rate is equal to zero; otherwise, the
release rate is provided by the detector data. If no detector data is available, then the release rate is set to the saturation flow
rate if a queue exists:
XX
rUi;t ¼ j k
mj;k; Ui;t  cUi;t ð8Þ

where
cUi;t = release rate of vehicles in lane group k of approach j (veh/s).
Eq. (9) represents the total delay from approaches during a specified time interval:
X X
AD ¼ Ui t
DUi ;t ð9Þ

where
AD = approach delay of one cycle (s).
3.2. Intersection delay

Intersection delay is calculated by tracking each turning movement and comparing the passage time of the vehicle with
the required minimum travel time. To obtain turning movement delay, AVDET uses the detector configuration and detection
data obtained from the Automatic Turning Movement Identification System (ATMIS) to estimate the intersection delay.
The algorithm used in ATMIS uses information collected from the detectors and the signal system of the intersection. By
tracking each detector’s status and the signal timing information second by second, the vehicle turning movements can be
estimated. To illustrate the algorithm, consider a four-legged intersection as shown in Fig. 5. Since there is only one lane per
direction at each intersection leg, the same lane is shared by left turning, right turning, and through-going vehicles. Two
types of detectors are used to determine the vehicle turning movements: the input detectors (shown in white in Fig. 5)
detect vehicles approaching the intersection, while the output detectors (shown in gray) detect vehicles leaving the
intersection.
For each turning movement, there is a fixed detector pair to be tracked by the ATMIS algorithm. For example, any north-
bound left-turning vehicle will first activate Detector 1 and then Detector 4, if the current signal allows for northbound left
turns. Similarly, every turning movement throughout the intersection can be defined by a sequence of detections, as shown
in Table 1. When fully coordinated with the signal timing plan, the algorithm is used to determine vehicle turning move-
ments from a specific detection sequence according to the data collected by each detector. More detailed information can
be found in (Xu et al., 2013) for interested researchers.
We can denote the distance between the input detectors to the output detectors as Lk and the approach speed limit as SL.
The minimum travel time required to pass through the distance of a movement Lk with speed SL can be calculated using Eq.
(10).
 I. Shatnawi et al. / International Journal of Transportation Science and Technology 7 (2018) 137–150 143

Fig. 5. Typical detector layout for a four-leg intersection (one-lane shared movements).

Lk
minðTT OI Þ ¼ ð10Þ
SL
By obtaining the timestamps (arrival and departure signatures) for vehicles crossing the intersection, the intersection
delay for each vehicle turning movement can be calculated as follows:

X X X TT OI 
Average intersection delay ¼ Ui j k N
 minðTT OI Þ ð11Þ
OI

where
I is the timestamp for the arriving vehicle,
O is the timestamp for the leaving vehicle,
N OI is the number of vehicles turned from I to O for lane group k,
TT OI is the measured travel time from I to O for all vehicles in the lane group, and
minðTT OI Þ is the lane group based on the minimal travel time between I and O.

4. Algorithm implementation

4.1. Input detection matching module

The input detection matching module is triggered by the input detectors. When the status of an input detector changes
from activated to deactivated, this indicates that a vehicle has entered the system. The detector’s identification number and

Table 1
Turning movements.

Input detector Output detector Movement Abbreviation

1 7 Northbound Through NT
1 6 Northbound Right Turn NR
1 8 Northbound Left Turn NL
2 8 Westbound Through WT
2 7 Westbound Right Turn WR
2 5 Westbound Left Turn WL
3 5 Southbound Through ST
3 8 Southbound Right Turn SR
3 6 Southbound Left Turn SL
4 6 Eastbound Through ET
4 5 Eastbound Right Turn ER
4 7 Eastbound Left Turn EL
144 I. Shatnawi et al. / International Journal of Transportation Science and Technology 7 (2018) 137–150

the deactivation timestamp will be recorded and saved on the waiting list. The records saved of the waiting list will be used
in the other two modules.

4.2. Output detection matching module

The output detection matching module is triggered by the output detectors. An output detection indicates that a vehicle
has left the intersection; there should be one and only one input detection matched with it. However, in practice, it will not
always be the case that a single input detection will be returned from the waiting list. As shown in Fig. 6, there are three
possible cases that need to be considered:

Fig. 6. Example of a case with multiple matches.

 I. Shatnawi et al. / International Journal of Transportation Science and Technology 7 (2018) 137–150 145

Case I: There is no matched detection on the waiting list: This situation can result from either a missed detection by input
detectors or a false detection from output detectors. Since any future detection will not be helpful for solving this prob-
lem, the system will not output anything except to mark the output detection as an error.

Case II: There is only one matched detection in the waiting list: This is the best condition for detection match, as it only
requires a simple search of the turning movement table to find the corresponding movement based on the detection pair.
The system will output this movement as the result and will mark the input and output detection as identified.

Case III: There is more than one matched detection on the waiting list: This is the most difficult situation that needs to be
considered in the algorithm; it can result from a number of reasons, such as an error in the detection of input/output
detectors or vehicles that have not cleared the intersection area. To illustrate the process for the algorithm with more
than one detection pair candidate, a simple example is provided in the following paragraph.

Consider a four-leg intersection with one lane in each direction as shown in Fig. 6, where a vehicle is moving from south
to east (i.e., a northbound right turn) and another vehicle is moving from north to south (i.e., a southbound through-going
vehicle). At any moment, the vehicles’ will have positions as shown in Fig. 6(a). Two candidate input detections are available
for the output detection. The system will hold all detections and not formulate a decision until the southbound vehicle leaves
the intersection. As shown in Fig. 6(b), there are now two output detections: one from Detector 5 and the other from Detec-
tor 6. Although there are still two possible matches for the output from Detector 6, there is only one input detection to match
with the output from Detector 5. The algorithm will output the movement as a southbound through-going vehicle. Conse-
quently, only one input detection will match with output Detector 6, and the algorithm will then output a northbound right
turn. Thus, the output detections are matched, and the turning movements are determined.

4.3. Input detection cleanup module

The input detection cleanup module is an independent process which cleans up the expired input detections in the detec-
tion waiting list. Input detections that have no matched output detection for a given time should be removed from the wait-
ing list in order to keep the system working properly. These non-matched detections may result either from false detections
indicated by input detectors or actual movements missed by output detectors. The expiration time is preset by the system
(typically, the length of time required for a signal phase), and the cleanup process is performed every second. This module of
the algorithm is crucial for helping to resolve cases with multiple matches.

4.4. Initial queue size

The initial queue size can be defined as the queue size at the beginning of each interval. When the system is first initialized,
the queue size is set to zero. As the simulation system progresses, the queue size will be tracked second by second.
Specifically, to calculate the approach delay, arrival detectors are placed in the approach to each intersection at a location
approximately 300 feet from the stop bar, while departure detectors are placed near the stop bar located ahead of the inter-
section, as shown in Fig. 8. For intersections with heavy traffic volume, arrival detectors are placed farther upstream (as far as
700 feet ahead of the intersection) to avoid queue spillback; tests have shown this distance is insensitive, as oversaturated
conditions are not considered to avoid street gridlock. The algorithm records vehicle arrivals second by second, and the data
are used to determine the arrival rates and departure rates of vehicles for use in Eqs. (7) and (8) in each lane group. The algo-
rithm then uses the vehicle arrival/departure data in each time interval of phase Ui to determine the delays experienced by
newly arriving vehicles as well as vehicles remaining in the queue based on the ‘‘first in/first out” rule.

Table 2
Corresponding turning movement for detector pair layout.

Arrival detector Departure detector Turning movement Abbreviation

1 20 Northbound Right Turn NBR
1 15 Northbound Through NBT
2 14 Northbound Through NBT
3 9 Northbound Left Turn NBL
6 5 Westbound Right Turn WBR
6 20 Westbound Through WBT
7 19 Westbound Through WBT
8 14 Westbound Left Turn WBL
11 10 Southbound Right Turn SBR
11 5 Southbound Through SBT
12 4 Southbound Through SBT
13 19 Southbound Left Turn SBL
16 15 Eastbound Right Turn EBR
16 10 Eastbound Through EBT
17 9 Eastbound Through EBT
18 4 Eastbound Left Turn EBL
146 I. Shatnawi et al. / International Journal of Transportation Science and Technology 7 (2018) 137–150

For measuring the intersection delay when a vehicle crosses the intersection, the algorithm uses the detector layout infor-
mation listed in Table 2 and shown in Fig. 8, along with the detection information and signal status, to identify a vehicle’s
turning movements and the corresponding delay using Eqs. (9) and (11). To help explain how to implement the proposed
methodology in real time, a flow chart of the work process of the AVDET algorithm is provided in Fig. 7.

Fig. 7. Flow chart of AVDET and data processing algorithm.

 I. Shatnawi et al. / International Journal of Transportation Science and Technology 7 (2018) 137–150 147

5. Experimental program

The performance of the AVDET algorithm was evaluated by conducting a microscopic simulation using PTV Vissim soft-
ware, in which a signalized intersection was built and calibrated to reflect real-world geometric and traffic parameters. To
investigate the feasibility and examine the robustness of the AVDET algorithm with variations in traffic volumes and per-
centages of turning movements, the simulation was run repetitively, with one hour for each scenario, using different random
seeds for the same set of parameters.
The signalized intersection selected for the analysis is shown in Fig. 8. The traffic volume includes 400, 800, 1200, and
1800 vehicle/hour/approach, and the turning movement percentages are 50%, 60%, 70% and 80% of the through volume.
The intersection approach speed limit was determined as 35 mph. Table 3 shows the details of the turning movements at
each demand level scenario.
Since Vissim internally tracks each vehicle’s movement in the simulation, it can measure the vehicle delays exactly.
Therefore, for comparison purposes, the delay outputs obtained directly from Vissim are considered as ‘‘ground-truth” val-
ues. A total of sixteen simulation cases were conducted.

6. Results and analysis

Tables 4–7 present the results of intersection control delay for all sixteen test cases where simulation was performed
using four volume demand levels (low, medium, high and saturated) and four turning volume combinations. Fig. 9 shows
a comparison of the results from the proposed algorithm and the Vissim simulation. As can be seen in this figure, the
root–mean–square errors (RMSE) for the average control delay are less than 1.0 s for a low traffic volume. The errors grad-
ually increase to 1.34, 2.4, and 6.7 s as the traffic volume increases to medium, high, and saturated, respectively. It can also be
observed that, at each level of traffic volume, the higher the through movement percentage (varying from 50% to 80%), the
smaller the errors.
It is clear that the AVDET method yields comparable results to those from Vissim. The error patterns demonstrate that an
crowded traffic volume and a higher volume of left-turning and right-turning vehicles can impact the accuracy of the algo-
rithm. However, the errors are reasonably limited for most scenarios where the total turning movement volume is under
50%.

6.1. Validation method: chi-square test

A commonly used method used to compare one series of data against another is the chi-square test, also referred to as v2
test. The chi-square will aid in determining whether there is a significant difference between the expected output generated
from Vissim and the output generated from the proposed AVDET algorithm in one or more intervals. If the numbers of indi-
viduals falling into each interval differs significantly from the number expected, the chi-square test can indicate whether the
difference between the expected and observed results is a significant difference due to sampling variation.
The general formula of the chi-square is presented in Eq. (12).

Fig. 8. Detector layout for AVDET.

148 I. Shatnawi et al. / International Journal of Transportation Science and Technology 7 (2018) 137–150

Table 3
Traffic volume and turning percentage variations.

Total approach volume (veh/h) 400 800 1200 1800

Scenario 1 Through Volume 80% 320 640 960 1440
Right-Turn Volume 10% 40 80 120 180
Left-Turn Volume 10% 40 80 120 180
Scenario 2 Through Volume 70% 280 560 840 1260
Right-Turn Volume 15% 60 120 180 270
Left-Turn Volume 15% 60 120 180 270
Scenario 3 Through Volume 60% 240 480 720 1080
Right-Turn Volume 20% 80 160 240 360
Left-Turn Volume 20% 80 160 240 360
Scenario 4 Through Volume 50% 200 400 600 900
Right-Turn Volume 25% 100 200 300 450
Left-Turn Volume 25% 100 200 300 450

Table 4
Delay results and chi-square analysis at 400 veh/hr approach volume and varying turning movement percentages.

Approach demand (veh/h) 400

Percentages of turning 50% Th, 25% L, 25% R 60% Th, 20% L, 20% R 70% Th, 15% L, 15% R 80% Th, 10% L, 10% R
movements (veh/h)
Time interval (s) Vissim AVDET Vissim AVDET Vissim AVDET Vissim AVDET
0–300 12.04 12.81 13.49 14.45 16.58 15.75 13.98 13.31
300–600 13.52 12.62 15.02 14.13 15.8 16.76 15.18 14.85
600–900 12.57 11.59 15.42 14.61 14.7 15.45 14.21 13.90
900–1200 11.8 12.87 14.07 15.00 14.35 15.32 14.2 13.84
1200–1500 13.56 14.55 15.11 14.35 14.83 13.86 14.39 14.18
1500–1800 12.23 13.25 14.85 15.69 15.22 16.20 13.82 14.65
1800–2100 13.34 12.52 15.56 16.24 15.34 16.10 15.11 13.97
2100–2400 13.03 12.01 14.92 13.88 15.24 14.46 15.13 14.49
2400–2700 13.12 12.21 14.49 15.18 14.8 14.30 14.22 13.38
2700–3000 12.86 11.96 14.33 15.36 14.5 15.41 14.94 13.52
3000–3300 13.01 13.78 14.35 15.35 15.11 16.44 14.93 14.57
3300–3600 12.38 11.49 13.8 14.82 14.2 14.53 13.3 13.85
Chi-square calculation df = 11 v211,0.05 = 0.81 df = 11 v211,0.05 = 0.67 df = 11 v211,0.05 = 0.61 df = 11 v211,0.05 = 0.44
a = 0.05 0.81 < 19.675 a = 0.05 0.67 < 19.675 a = 0.05 0.61 < 19.675 a = 0.05 0.44 < 19.675
Not reject Ho Not reject Ho Not reject Ho Not reject Ho

Table 5
Delay results and chi-square analysis at 800 veh/hr approach volume and varying turning movement percentages.

Approach demand (veh/h) 800

Percentages of turning 50% Th, 25% L, 25% R 60% Th, 20% L, 20% R 70% Th, 15% L, 15% R 80% Th, 10% L, 10% R
movements (veh/h)
Time interval (s) Vissim AVDET Vissim AVDET Vissim AVDET Vissim AVDET
0–300 21.56 20.15 21.16 22.73 21.68 20.65 20.61 19.23
300–600 20.02 18.88 19.16 20.77 19.21 18.11 18.36 19.43
600–900 18.63 20.16 18.39 19.53 18.28 19.85 17.92 16.78
900–1200 19.46 17.81 19.26 20.53 19.41 20.22 19.28 18.38
1200–1500 18.58 17.29 18.89 19.85 18.77 19.75 18.24 17.12
1500–1800 18.05 19.32 18.46 17.21 18.07 19.24 16.83 17.87
1800–2100 19.24 18.15 19.39 20.76 19.93 18.51 18.79 17.62
2100–2400 19.02 17.57 19.55 20.80 19.4 20.61 18.73 17.76
2400–2700 19.71 20.83 19.69 18.44 18.89 20.09 19.12 18.12
2700–3000 18.13 19.27 17.51 18.82 18.58 17.40 17.66 18.81
3000–3300 18.21 19.43 18.38 19.40 17.86 19.11 17.06 17.92
3300–3600 19.66 21.32 20.41 21.68 20.04 18.73 19.19 18.21
Chi-square calculation df = 11 v211,0.05₅ = 1.13 df = 11 v211,0.05 = 1.03 df = 11 v211,0.05 = 0.91 df = 11 v211,0.05 = 0.75
a = 0.05 1.13 < 19.675 a = 0.05 1.03 < 19.675 a = 0.05 0.91 < 19.675 a = 0.05 0.75 < 19.675
Not reject Ho Not reject Ho Not reject Ho Not reject Ho
 I. Shatnawi et al. / International Journal of Transportation Science and Technology 7 (2018) 137–150 149

Table 6
Delay results and chi-square analysis at 1200 veh/hr approach volume and varying turning movement percentages.

Approach demand (veh/h) 1200

Percentages of turning 50% Th, 25% L, 25% R 60% Th, 20% L, 20% R 70% Th, 15% L, 15% R 80% Th, 10% L, 10% R
movements (veh/h)
Time interval (s) Vissim AVDET Vissim AVDET Vissim AVDET Vissim AVDET
0–300 31.44 33.25 30.23 31.93 34.14 35.92 29.26 30.54
300–600 40.62 38.33 36.24 38.01 37.33 39.10 33.49 32.26
600–900 48.48 46.18 42.92 45.11 42.78 41.20 37.21 38.68
900–1200 50.76 52.74 41.95 44.21 43.15 40.38 36.31 38.16
1200–1500 51.9 49.38 41.92 44.04 40.44 42.53 36.44 34.75
1500–1800 49.66 46.76 38.81 40.43 43.47 45.57 33.55 32.22
1800–2100 56.57 53.81 44.69 43.12 43.87 41.58 34.31 33.01
2100–2400 55.7 58.15 42.68 40.65 44.26 42.01 36.72 38.53
2400–2700 56.32 54.19 44.44 46.45 46.66 44.46 35.61 34.10
2700–3000 51.23 53.72 46.36 48.81 44.18 42.07 34.12 35.52
3000–3300 54.54 52.08 48.45 45.12 42.01 43.69 35.41 33.98
3300–3600 53.92 56.36 44.48 46.36 39.47 41.09 35.38 37.87
Chi-square calculation df = 11 v211,0.05 = 1.38 df = 11 v211,0.05 = 1.27 df = 11 v211,0.05 = 1.2 df = 11 v211,0.05 = 0.88
a = 0.05 1.38 < 19.675 a = 0.05 1.27 < 19.675 a = 0.05 1.2 < 19.675 a = 0.05 0.88 < 19.675
Not reject Ho Not reject Ho Not reject Ho Not reject Ho

Table 7
Delay results and chi-square analysis at 1800 veh/hr approach volume and varying turning movement percentages.

Approach demand (veh/h) 1800

Percentages of turning 50% Th, 25% L, 25% R 60% Th, 20% L, 20% R 70% Th, 15% L, 15% R 80% Th, 10% L, 10% R
Movements (veh/h)
Time interval (s) Vissim AVDET Vissim AVDET Vissim AVDET Vissim AVDET
0–300 113.56 120.43 118.87 126.90 111.93 117.19 103.99 108.92
300–600 111.31 118.18 113.94 108.28 113.35 118.59 101.65 107.27
600–900 114.33 121.15 119.38 126.44 114.32 108.67 108.63 102.51
900–1200 117.25 110.80 120.01 115.32 111.09 116.02 109.36 104.48
1200–1500 115.11 107.15 115.29 123.16 119.35 113.62 105.82 100.85
1500–1800 114.66 108.13 119.95 125.88 112.05 106.90 108.67 113.10
1800–2100 115.24 120.93 116.02 123.03 115.13 109.22 105.93 100.72
2100–2400 112.23 119.66 117.38 111.88 112.53 107.21 104.26 109.86
2400–2700 117.64 110.83 120.06 124.73 118.82 113.22 106.39 101.35
2700–3000 118.49 112.51 118.85 113.19 118.28 112.96 106.31 110.20
3000–3300 117.96 110.92 120.98 126.11 116.77 122.07 106.36 101.80
3300–3600 116.78 122.83 125.04 117.14 113.98 120.13 109.58 103.81
Chi-square calculation df = 11 v211,0.05 = 4.73 df = 11 v211,0.05 = 4.1 df = 11 v211,0.05 = 3.13 df = 11 v211,0.05 = 2.95
a = 0.05 4.73 < 19.675 a = 0.05 4.1 < 19.675 a = 0.05 3.13 < 19.675 a = 0.05 2.95 < 19.675
Not reject Ho Not reject Ho Not reject Ho Not reject Ho

Fig. 9. RMSE results at varying approach volumes and turning movement percentages.
150 I. Shatnawi et al. / International Journal of Transportation Science and Technology 7 (2018) 137–150

Xn ðO  EÞ2
v2 ¼ ð12Þ
i¼1 E
where

v2 is the value for chi-square,

i is the interval number,
O is the observed frequency, and
E is the expected frequency.

In order to examine whether there are statistically significant differences between the output delays from AVDET and Vis-
sim, a chi-square test was conducted on the results under varying traffic conditions, the results are presented in Tables 4–7.
The chi-square test with the level of significance a = 0.05 is used to test the null hypothesis. Since there are 12 intervals of
categories in the simulation test, the degree of freedom (df) becomes df = 12–1 = 11. Checking the chi-square from the dis-
tribution table with a = 0.05 and df = 11 shows that v211,0.05 = 19.675. Since the calculated values for chi-square are less than
the chi-square tabulated value, we cannot reject the null hypothesis Ho, which means that there is no significant difference
between the estimated delays from AVDET and those from Vissim.

7. Conclusions and future work

Vehicle delay is one the most important performance measures of effectiveness (PMOE), as it allows traffic engineers to
evaluate the performance of a signal system. In this paper, we presented and tested the AVDET algorithm for automatically
measuring intersection control delay. Obtaining vehicle delay in real-time allows traffic engineers to evaluate the perfor-
mance of a signal system and make the necessary adjustments in the signal timing plans to reduce the overall intersection
delay. Validation through simulation has shown the ability of AVDET to estimate the vehicle delay with reasonable accu-
rately by incorporating the real-time detector data. In our extensive lab tests, AVDET was found to perform very effectively
under different traffic scenarios (low, medium, high, and saturated); oversaturated conditions which may cause street grid-
lock have not considered in this study. The methodology developed in this study was applied to an isolated intersection; in
the next phase of the ongoing research, the methodology will be expanded to large-scale data applications to estimate vehi-
cle delays for a traffic corridor and over an entire traffic network.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.ijtst.2018.
04.001.

References

Abdel-Rahim, A., Dixon, M.P., 2009. Automated measurement of approach delay at signalized intersections: vehicle-event-based method Paper no. 09-1287.
TRB 88th Annual Meeting Compendium of Papers DVD. Transportation Research Board (TRB), National Research Council, Washington D.C..
Forbush, T.R., 2011. Automated Delay Estimation at Signalized Intersections: Phase I Concept and Algorithm Development (Master’s thesis) Retrieved from.
All Theses and Dissertations, p. 2471.
Highway Capacity Manual (HCM), 2010. In: Interrupted Flow. Transportation Research Board, Washington, D.C.. https://doi.org/10.5555/
9780309160773.26.
Kebab, W., Dixon, M., Abdel-Rahim, A., 2007. Field measurement of approach delay at signalized intersections using point data. Transp. Res. Rec., 37–44
Kinzel, C.D., 1992. Traffic studies. In: Pine, J.L. (Ed.), Traffic Engineering Handbook. Prentice-Hall, Upper Saddle River, N.J.. 4th ed..
Mousa, R.M., 2002. Analysis and modeling of measured delays at isolated signalized intersections. J. Transp. Eng. 128 (4), 347–354.
Quiroga, C.A., Bullock, D., 1999. Measuring control delay at signalized intersections. J. Transp. Eng. 125 (4), 271–280.
Teply, S., 1989. Accuracy of delay surveys at signalized intersections. Transp. Res. Rec. 1225, 24–32.
Xu, K., Yi, P., Shao, C., Mao, J., 2013. Development and testing of an automatic turning movement identification system at signalized intersections. J. Transp.
Technol. 2013, 241–246.

Further reading

Committee on Operating Speeds in Urban Areas, 1955. Speed and Travel Time Measurement in Urban Areas. Highway Research Board, Washington, D.C..