A Study of the Travel Time of Intersections by Vehicles using

Computer Vision

V. D. Shepelev

1,2 a

, A. I. Glushkov

, Z. V. Almetova

and V. G. Mavrin

South Ural State University, 76 Lenin Avenue, Chelyabinsk, Russia

Silkway International University, 27 “А” Tokaev Street, Shymkent, Kazakhstan

Kazan Federal University, Suyumbike Avenue, 10A, Naberezhnue Chelny, Russia

Keywords: Vehicle Queue, Crossing the Intersections, Travel Time at Intersections, Statistical Confidence, Intelligent

Transport Systems.

Abstract: The article deals with the problem of intelligent traffic control at intersections of road networks of large cities.

Due to the advances in cyber-physical systems (CPS), autonomous driving, as part of Intelligent Transport

Systems (ITS), will obviously be in the centre of future urban transport. However, the existing ITSs do not

fully take into account the size, structure, and parameters of the queue of vehicles waiting at inter-sections,

which in turn affects the traffic capacity of the intersection. In the study, we used computer vision to interpret

a queue of vehicles and record the parameters at the intersection on a real time basis. We studied the mutual

impact of two generalized categories of transport standing in the queue before the stop line at the intersection.

We developed a general conceptual research model, which includes both the task of forming the original

sample and statistical analysis of the time needed to cross an intersection by the vehicles located in different

initial positions. The main research results showed a statistically significant reduction in the vehicle speed to

58% in case there are various categories of vehicles standing in the queue at the intersection.

1 INTRODUCTION

Currently, researchers pay much attention to the

emerging advanced intelligent transport systems,

such as vehicle-to-vehicle communications and the

vehicle-to-infrastructure communication system

(Azimi et al., 2015; Arsava et al., 2014). Using the

advantages of such systems, scientists develop a joint

control mod-el which optimizes the speeds of

connected vehicles and simultaneously coordinates

signals along the artery. This control model divides

the connected vehicles into groups so that the vehicles

can cross intersections together without stops or with

the least stop time. At the same time, it optimizes

signal synchronization patterns along the artery to

achieve lower signal delays and a higher capacity

(Wang et al., 2020; Bakibillah et al., 2019; Makarova

et al., 2019).

https://orcid.org/0000-0002-1143-2031

https://orcid.org/0000-0002-6292-7122

https://orcid.org/0000-0001-9304-8406

https://orcid.org/0000-0001-6681-5489

To solve the problem of traffic jams at road

intersections through intelligent systems, it is

necessary to form a comprehensive insight into

several determining problem factors. The most

critical factor in the tasks of setting and modelling

intelligent transport systems is the queue of vehicles

waiting at the intersection.

Many studies deal with the development of

adaptive learning controllers for traffic signals,

analyse their studied policies, and compare them with

the Webster controller. Using the representation of

the state of statistics (i.e., vehicle queues and density),

the proposed traffic signal controllers with enhanced

training develop an optimal policy in the dynamic and

stochastic traffic microsimulation (Tubaishat et al.,

2008; Genders et al., 2020). Ghazal et al. (2016) and

Mandal et al. (2020) offer a system based on the use

of microcontrollers which assess the traffic density

using IR sensors.

Shepelev, V., Glushkov, A., Almetova, Z. and Mavrin, V.

A Study of the Travel Time of Intersections by Vehicles using Computer Vision.

DOI: 10.5220/0009806206530658

In Proceedings of the 6th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2020), pages 653-658

ISBN: 978-989-758-419-0

653

The study Kulakarni et al. (2020) proposes a

model between the saturation flow and the ratio of a

small car, a large car and the road width.

There is a widely used method based on the

formulation of the combinatorial problem to

structurally optimize the traffic light cycle, which

allows us to create the most optimal phase separation

schemes according to the set traffic demand (Galkin

et al., 2018; Kapski et al., 2019; Kazhaev et al., 2018).

Several studies propose a behavioural view

methodology based on the calibration of a

microsimulation model for highly heterogeneous

traffic at the signalized intersection. The proposed

methodology is illustrated through the use of Verkehr

in Staedten Simulation - a widely used

microsimulation soft-ware based on the psycho-

physical car tracking model. Signalized intersections

with different transport and geometric characteristics

from two cities of India were taken as an example

(Mathew et al., 2010).

In (Treiber et al., 2013), the main part, after

presenting various categories of traffic data, deals

with the mathematical description of the dynamics of

traffic flows, covers macroscopic models describing

traffic in terms of density, as well as microscopic

models of many particles, in which each particle

corresponds to a vehicle and its driver. Since the

stochastic queue model excludes some realistic traffic

flow aspects, the authors propose to adapt maximum

pressure control for dynamic traffic assignment based

on modelling (Levin et al., 2019)

Some studies dealing with the traffic instability

and calibration / validation of models present these

topics in a new and systematic way (Solovyev et

al.,2020; Makarova et al., 2018; Ertman et al., 2016).

Quek et al. (2006) describes the use of a particular

class of a fuzzy neural network, known as a fuzzy

neural network of a pseudo-external product, using

the truth value restriction method (POPFNN-TVR)

for a short-term forecast of the traffic flow.

Considering the effects of queues, (Wu et al.,

2020; Dai et al., 2016) offered a method for defining

assessment and classification criteria for a

supersaturated state based on the distance between the

intersections, traffic demand, and intersection

capacity.

A study of Kikuzawa et al. (2019) developed a

traffic flow measurement system to extract the traffic

data by analysing images from fixed-point cameras

installed near intersections. The intersection and the

speed of recognizing the correspondence for the same

vehicle and the accuracy of the average speed are

estimated by the difference in the time needed to cross

two intersections.

There are a minimal number of studies in the field

of investigating the mutual impact of various

categories of vehicles in the queue on the capacity of

signalized intersections. The most representative

works are summarized in the introduction.

The closest study, which describes the headway,

speed and acceleration characteristics of vehicles

during the queue discharge after the green onset under

mixed traffic conditions, is presented by Dey et al.

(2013). It has been found that the queue discharge

headway shows an unmistakable pattern of gradual

compression as queuing vehicles are discharged in

succession. This article also analyses the speed at

which vehicles move when the queue is discharged

for two different categories of vehicles (category I -

passenger cars; category 2 - trucks and buses). It also

analyses the acceleration characteristics of different

categories of vehicles released from a stop after the

green onset and explains them by the non-uniform

acceleration model in the form of. A probabilistic

approach based on the second-order first-moment

method was adopted to assess the saturation flow and

delay caused by the traffic at signalized intersections

under non-uniform traffic conditions (Thamizh et al.,

1995).

The presented methodologies available for

assessing the capacity of signalized intersections are

based on the concept of saturation flow(s). At the

same time, the studies do not fully take into account

the impact of such factors as the queue structure and

its dynamic parameters on the capacity of inter-

sections. In this paper, we will focus on these two

aspects.

2 METHODS

The purpose of this article is to analyse the impact of

heterogeneity of vehicles in the queue on the capacity

of intersections using computer vision. To achieve

this objective, let us divide the problem into three

sub-tasks: to determine how the time needed to cross

the intersection crossing time decreases due to the

traffic-light stop, to analyse the reduction in the time

needed to cross the intersection due to the vehicle stop

in the queue, and to determine the statistical

significance of changes in the timed needed to cross

the intersection crossing time. In the following

sections, we will describe in detail each module,

together with the data collected for analysis and

assessment.

iMLTrans 2020 - Special Session on Intelligent Mobility, Logistics and Transport

654

2.1 Dataset

Street surveillance cameras were used to collect data,

which provide a stable 25 frames per second,

supporting a resolution of 1920

× 1080 (Fedorov et

al., 2019).

As shown in Figure 1, images from these cameras

differ from those in publicly available data sets, such

as KITTI (Cordts et al., 2016) or UA-DETRAC (Wen

et al., 2015).

Figure 1: An example of an image from CCTV cameras.

To carry out our research, we focused on one

camera, which controls one of the most problematic

intersections in the city of Chelyabinsk. It is a

conscious choice to maximize detection accuracy and

obtain a minimally viable product.

The data on the time the vehicles needed to cross

the stop-line section, and the intersection were

collected through the use of the YOLOv3 trained

neural network (Figure 2).

Stage 1. To determine how the time needed to

cross the intersection decreases due to the vehicle

stop in the queue.

The general conceptual research model can be

presented by the following interdependent stages:

1. Adjustment of the input data to make them

uniform;

2. Analysis of the average values of the time

needed to cross the intersection by the vehicles

located in different initial positions;

3. Determination of the statistical significance of

changes in the time needed to cross the intersection

due to the traffic-light stop.

The formation of a homogeneous sample.

To minimize distortions of the research results, we

selected a sample of the intersection crossing by

similar vehicles. For the analysed intersection, these

are vehicles of category M - vehicles with at least four

wheels and used to carry passengers (Vehicle

Categories, 2019). We also rejected the observations

with the vehicles of category M3 (which maximum

permissible mass exceeds 5 tons), which will also

bring distortions into the results due to their size and

response rate.

Figure 2: An example of an image from our dataset.

We also rejected the observations falling out of

the general intersection crossing structure - which

have a short queue without vehicles crossing the

intersection non-stop. The initial sample was reduced

by 38% and amounted to 1128 observations for the

vehicles of categories M1 and M2. A visual

illustration of the empirical data accepted for our

analysis is shown in Figure 3.

Stage 2. To analyse the decrease in the time

needed to cross the intersection due to the vehicle stop

in the queue.

In the prepared homogeneous sample of the

observations of vehicles crossing the intersection, we

also determined the time intervals when the vehicles

entered the intersection, which will allow us to solve

similar research problems further. The primary

information on this study – the time needed to cross

the intersection average for the sample - is divided

into four positions according to the location of a

vehicle and presented in Table 1.

Table 1: Changes in the time needed to cross the

intersection.

Position of the

vehicle

Time needed to

cross the

intersection,

seconds

Reduction in

speed, %

The first vehicle

in the

ueue

6.56 57.8

The last vehicle

in the queue

5.25 67.3

The first vehicle

assin

non-sto

4.62 76.5

The last vehicle

assin

non-sto

3.53 100

As follows from the calculations, the speed of the

vehicle crossing a free intersection decreases by

almost a half (up to 58%), depending on the presence

of the queue. However, to confirm the reliability of

the obtained results, it is necessary to assess the

statistical significance of their differences.

A Study of the Travel Time of Intersections by Vehicles using Computer Vision

655

Figure 3: The empirical data on the time needed to cross the intersection.

Stage 3. To determine the statistical significance

of changes in the time needed to cross the inter-

section

In a statistical analysis, a manifested regularity is

considered to be statistically significant if the

empirical significance level is less than the generally

accepted critical value of 0.05 (5%) (Buyul et al.,

2005; Tyurin et al., 2016).

For the problem under consideration – to assess

the statistical reliability of reducing the time needed

to cross the intersection in the presence of a queue, it

is necessary to make sure that the calculated average

time values (Table 1) in terms of statistics are not in

the same confidence interval, i.e., do not represent

one and the same numerical value.

To select a statistical analysis method most

suitable in this case, it is necessary to determine its

following characteristics:

 This is a parametric method since the data

measurement scale is at least interval; the

distribution law is close to normal (not

verified);

 The samples are linked since each quadruple of

data in one record is read from one observation.

Given such restrictions, the most acceptable is the

method for two linked samples - Student’s parametric

t-test, which allows us to compare each pair of four

calculated time values. Otherwise, we can use the

parametric method for several linked samples (>2) -

an analysis of variance with repeated measurements,

which allows us to evaluate the differences between

all the four values simultaneously. A detailed analysis

of all pairs of time values is undoubtedly of great

interest, so we select the first method.

The statistics calculated by Student’s t-test is

determined as follows:







2121









)

where

, xx

is the mathematical expectation for the

two samples; μ

, μ

are the estimates of the average

values of the two samples (it is assumed that μ

=μ

);



is the normalization base formed from the

estimates of variances S of these equal samples (n

= n =128) according to the standard rule.









)

Based on the calculated values of these statistics,

we determined the empirical significance levels for

all pairs of average time values (Table 1), which are

summarized, together with Pearson’s paired

correlation coefficients, in Table 2.

Comparing the calculated values of the error

(these are the selected cells “Difference estimate” in

Table 2) with an acceptable level of 5% (5.00E-2), we

made the following conclusion - all pairs of the

average values of the time needed to cross the

intersection (Table 1) differ significantly. I.e., in

terms of statistics - the differences between them are

statistically significant.

To further confirm the legitimacy of the obtained

result, we determined paired correlation coefficients

and their corresponding levels of statistical

confidence (Table 2). All the links relate to the levels

of weak and very weak links, which is confirmed by

the absence of hidden regularities in the empirical

data, which could introduce distortions into the

analysis results.

3 CONCLUSIONS

All the models are based on uniform driving

conditions; these models may not provide a

satisfactory estimate of delays under non-uniform

iMLTrans 2020 - Special Session on Intelligent Mobility, Logistics and Transport

656

Table 2: Changes in the time needed to cross the intersection.

The first

vehicle in the

ueue

The last

vehicle in the

ueue

The first

vehicle passing

non-sto

The last

vehicle passing

non-sto

The first vehicle

in the queue

Pearson’s correlation

Value (2 sides)

(

sam

le size

)

=128

The last vehicle

in the queue

Pearson’s correlation

Value (2 sides)

Difference estimate

0.233**

0.008

1.170E-20

=128

The first vehicle

passing non-stop

Pearson’s correlation

Value (2 sides)

Difference estimate

0.169

0.056

3.031E-28

0.360

0.000

2.066E-08

=128

The last vehicle

passing non-stop

Pearson’s correlation

Value (2 sides)

Difference estimate

0.073

0.414

3.097E-50

0.236**

0.007

3.308E-39

0.171

0.054

2.084E-16

=128

traffic conditions. The average delay caused by

various categories of vehicles at signalized

intersections is an essential criterion in assessing the

capacity of these intersections. The use of neural

networks in the tasks covering the recognition and

interpretation of traffic flows allows us to obtain the

new data needed to optimize intelligent transport

systems. As a result of the data analysis at the studied

intersection, we have found that the time needed to

cross the intersection by the vehicles of categories M1

and M2 reduces in a statistically significant manner

to 58% if there is a queue. The interpretation of a bulk

of the data in the vehicle-infrastructure system will

allow us to introduce advanced solutions for

autonomous transport, such as identification of

vehicle parameters and queuing at intersections to

increase their capacity.

ACKNOWLEDGEMENTS

This work was supported by the Russian Foundation

for Basic Research: grant No. 19-29-06008\19.

REFERENCES

Arsava, T., Xie, Y., Gartner, N. H., Mwakalonge, J., 2014.

Arterial traffic signal coordination utilizing vehicular

traffic origin-destination information. In Proceedings

of the 17th IEEE International Conference on

Intelligent Transportation Systems (ITSC), pages 2132-

2137. IEEE.

Azimi, R., Bhatia, G., Rajkumar, R., Mudalige, P., 2015.

Ballroom intersection protocol: Synchronous

autonomous driving at intersections. In Proceedings of

the IEEE 21st International Conference on Embedded

and Real-Time Computing Systems and Applications,

(RTCSA), pages 167-175.

Bakibillah, A. S. M., Hasan, M., Rahman, M. M., Kamal,

M. A. S. (2019). Predictive car-following scheme for

improving traffic flows on urban road networks.

Control Theory and Technology, 17(4): 325-334.

Buyul, A. Zöfel, P. 2005. SPSS: the art of information

processing. Analysis of statistical data and restoration

of hidden patterns, DiaSoft, Moscow.

Cordts, M., Omran M., Ramos, S., Rehfeld T., Enzweiler,

M., Benenson, R., Franke, U., Roth, S., Schiele, B.,

2016. The cityscapes dataset for semantic urban scene

understanding. In 2016 IEEE Conference on Computer

Vision and Pattern Recognition (CVPR), pages 3213–

3223. IEEE.

Dai, P., Liu, K., Zhuge, Q., Sha, E. H.., Lee, V. C. S., Son,

S. H., 2016. Quality-of-experience-oriented

autonomous intersection control in vehicular networks.

IEEE Transactions on Intelligent Transportation

Systems, 17(7): 1956-1967.

Dey, P. P., Nandal, S., Kalyan, R., 2013. Queue discharge

characteristics at signalised intersections under mixed

traffic conditions. European Transport - Trasporti

Europei, 55.

Ertman, S., Ertman, J., Zakharov, D. Adaptation of urban

roads to changing of transport demand. In Proceedings

of the International Conference on Sustainable Cities

(ICSC), 6(01013).

Fedorov, A., Nikolskaia, K., Ivanov, S., Shepelev, V.,

Minbaleev, A., 2019. Traffic flow estimation with data

from a video surveillance camera. Journal of Big Data,

6(1).

Galkin, A., Lobashov O., Capayova S., Hodakova D.,

Schlosser T., 2018. Perspective of decreasing of road

traffic pollution in the cities. In Proceedings of the 18th

International Multidisciplinary Scientific

Geoconference (SGEM), 4.2 (18): 547-554.

Genders, W., Razavi, S., 2020. Policy analysis of adaptive

traffic signal control using reinforcement learning.

Journal of Computing in Civil Engineering, 34(1).

Ghazal, B., Elkhatib, K., Chahine, K., Kherfan, M., 2016.

Smart traffic light control system. In Proceedings of the

A Study of the Travel Time of Intersections by Vehicles using Computer Vision

657

3rd International Conference on Electrical,

Electronics, Computer Engineering and their

Applications (EECEA), pages 140-145.

Kapski, D., Kasyanik, V., Lobashov, O., Volynets, A.,

Kaptsevich, O., Galkin, A., 2019. Estimating the

parameters of traffic flows on the basis of processing of

localization data on the movement of vehicles.

Communications-Scientific letters of the University of

Zilina, 21(2): 89-99.

Kazhaev, A., Almetova, Z., Shepelev, V., Shubenkova, K.,

2018. Modelling urban route transport network

parameters with traffic, demand and infrastructural

limitations being considered. In IOP Conference

Series: Earth and Environmental Science (ICSC),

177(1).

Kikuzawa, M., Jeong, M., 2019. Development of traffic

flow measurement system using fixed point cameras. In

ACM International Conference Proceeding Series,

pages 79-83.

Kulakarni, R., Chepuri, A., Arkatkar, S., Joshi, G. J., 2020.

Estimation of saturation flow at signalized intersections

under heterogeneous traffic conditions. Lecture Notes

in Civil Engineering, 45:591-605.

Levin, M. W., Rey, D., Schwartz, A., 2019. Max-pressure

control of dynamic lane reversal and autonomous

intersection management. Transportmetrica B, 7(1):

1693-1718.

Makarova, I., Shubenkova, K., Mavrin, V., Buyvol, P.,

2018. Improving safety on the crosswalks with the use

of fuzzy logic. Transport Problems, 13(1): 97-109.

Makarova, I., Yakupova, G., Buyvol, P., 2019. Improving

road safety by affecting negative factors. VEHITS -

Proceedings of the 5th International Conference on

Vehicle Technology and Intelligent Transport Systems.

2019. P.629-637.

Mandal, A., Sadhukhan, P., Gaji, F., Sharma, P., 2020.

Measuring real-time road traffic queue length: A

reliable approach using ultrasonic sensor. Lecture Notes

in Electrical Engineering, 602:391-398.

Mathew, T. V., Radhakrishnan, P., 2010. Calibration of

microsimulation models for nonlane-based

heterogeneous traffic at signalized intersections.

Journal of Urban Planning and Development, 136(1):

59-66.

Quek, C., Pasquier, M., Lim, B. B. S., 2006. Pop-traffic: A

novel fuzzy neural approach to road traffic analysis and

prediction. IEEE Transactions on Intelligent

Transportation Systems, 7(2): 133-146.

Solovyev, A. A., Valuev, A. M., 2020. Structural and

parametric control of a signalized intersection with real-

time “Education” of drivers. Advances in Intelligent

Systems and Computing, 902:517-526.

Thamizh, A., V., Jagadeesh, K., 1995. Effect of

heterogeneity of traffic on delay at signalized

intersections. Journal of Transportation Engineering,

121(5): 397-404.

Treiber, M., Kesting, A., 2013. Traffic flow dynamics:

Data, models and simulation. Springer Berlin

Heidelberg.

Tubaishat, M., Qi, Q., Shang, Y., Shi, H., 2008. Wireless

sensor-based traffic light control. In Proceedings of the

5th IEEE Consumer Communications and Networking

Conference, (CCNC), pages 702-706.

Tyurin, Yu.N., Makarov, A.A., 2016. Analysis of data on

the computer, MCCNMO, Moscow.

Vehicle Categories - M, N, O, L., 2019. Availible at

https://avtomirrf.ru/kategorii-ts.html.

Wang, P., Jiang, Y., Xiao, L., Zhao, Y., Li, Y., 2020. A joint

control model for connected vehicle platoon and arterial

signal coordination. Journal of Intelligent

Transportation Systems: Technology, Planning, and

Operations, 24(1): 81-92.

Wen L, Du D, Cai Z, Lei Z, Chang MC, Qi H, Lim J, Yang

MH, Lyu S., 2015. UA-DETRAC: a new benchmark

and protocol for multi-object detection and tracking.

arXiv:1511.04136.

Wu, L., Ci, Y., Wang, Y., Chen, P., 2020. Fuel consumption

at the oversaturated signalized intersection considering

queue effects: A case study in Harbin, China. Energy,

192.

iMLTrans 2020 - Special Session on Intelligent Mobility, Logistics and Transport

658