See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/357717170

Traffic Management using Machine Learning

Article · January 2022

CITATION READS
1 7,798

1 author:

Ritu Ritu
Bournemouth University
3 PUBLICATIONS 1 CITATION

SEE PROFILE

All content following this page was uploaded by Ritu Ritu on 10 January 2022.

The user has requested enhancement of the downloaded file.

 Traffic Management using Machine Learning

Ritu
Department of Computing
Bournemouth University
Poole United Kingdom
[email protected]

Abstract - Machine learning has multiple related to the working principle as well as
real-world applications which allow faster the challenges that can transpire in such a
and efficient functioning of programs and system are also described in this paper.
systems. This paper investigates the Proper recommendations are also provided
application of machine learning algorithms for developers to create an intelligent
within intelligent transport systems, transport system with the help of machine
specifically traffic management. With learning algorithms that not only properly
increasing traffic across the globe as a manage traffic data but also allow
result of the rising population and number supportive assessments and evaluation to
of vehicles, traffic management has modern traffic management systems.
increasingly become complex and requires
additional accuracy and efficiency apart Keywords – machine learning, Artificial
from human intervention. Machine Intelligence, Traffic Management System,
learning and its related algorithms can Rectangular Region of Interest, convolutional

fulfill these criteria as researched and neural network, You Only Look Once, Intelligent
Transportation System.
provided in this paper for further
consideration. As a result, multiple themes
I. Introduction the available traffic infrastructure. The
The study has helped in understanding how problem has been tried to be addressed with
traffic management can be done with help of help of the Traffic Management System
Machine Learning. Machine Learning has (TMS) and it monitors the vehicles to
been a field of Artificial Intelligence (AI) decrease the time spend in traffic signals
and it helps in the prediction of road traffic along with suggesting alternative routes to
as well as detection of the data. Machine the vehicles [6]. Traffic congestion has been
Learning has been defined as a part of AI prevented with help of TMS which works
which works with the help of a trained input based on the input of video sequence taken
model. Machine Learning Algorithms are from the convolutional neural network. The
formed based on a mathematical model and training process was implemented through
help in the creation of the Intelligent Traffic the help of convolutional neural network
Management System. In the current study, at topology made with help of the YOLO
first, the adaptation of traffic management Algorithm. The technique works through
systems has been explored with the help of spatial detection of the object in a video
Machine learning. Secondly, the study has frame. It has been considered as the input in
attempted to discuss the application of the most of the tracking algorithms. TMS using
YOLO algorithm and the proposed system. machine learning also makes use of
In this study, furthermore, the aspect of a Rectangular Region of Interest (ROI) and
smart traffic management system has been segments the objects [6]. The functionality
explored with machine learning. The of the Intelligent Traffic Signal Management
contribution of machine learning has been system with help of cloud vision API and
analyzed in traffic prediction. This study has Machine Learning has been described by
also discussed the relationship between Shwetha R.J and her co-authors in 2018.
machine learning and traffic congestion Images of traffic junction are taken and
management. update is made to the cloud through the help
of cloud vision API density.
II. Adaption of Traffic Management The type of vehicles in the traffic which has
System with help of Machine Learning to be detected are taken as input and update
Congestion has become a difficult issue as is made to the cloud through the help of
the network traffic has increased faster than cloud vision API density. The type of
vehicle is detected which in turn sends the Look Once) real-time object detection
status to the previous signal. The previous algorithm has been one of the most effective
signal which is the present one will check one for the detection of the object. It has
for the status of the next signal and proceed been observed that object detection is one of
with the operation [10]. This process helps the most important criteria or processes to be
in triggering the RFID placed next to the implemented within autonomous vehicle
signal and detects the vehicles crossing the technology. YOLO works based on a clever
platform. Similarly, there has been a convolutional neural network (CNN) and
proposal of an Adaptive Traffic with it, object detection is done effectively
Management System (ATMS) that makes [6]. The algorithm has been beneficial for
use of IoT and Machine Learning Camera application within the single neural network
Sensors along with two controller boards. and is applied to the full image. The image
Camera Sensors capture the details from the then gets divided into various regions and
lane and pass the data to the first controller the bounding boxes are predicted along with
board. The controller board uses this to the probabilities of each region [9]. The
adjust the traffic signals and congestion predicted probabilities are used for waiting
lights accordingly. Min-Max Fairness for the bounding boxes.
algorithm is used after estimating the With help of YOLO, the advantage that is
difference between the counts of the two obtained is of having a high accuracy while
lanes [6]. The priority is given to the lane being able to run in real-time. The algorithm
having a low average waiting time. works by looking at the object once and the
image is sensed while proceeding with the
III. Application of YOLO Algorithm and only one forward propagation. It passes
Proposed System through the neural network and makes
Machine Learning can be considered as one predictions [6]. It has been observed that
of the effective areas of application of with help of YOLO, single CNN
Artificial Intelligence (AI) in which a simultaneously predicts the multiple
computer system learns from previous bounding boxes and trains on complete
experiences. Machine Learning algorithm images. It helps in direct optimization of the
provides the advantage of improving the detection performance. The proposed system
process from the training. YOLO (You Only has been shown in this case in which the
input video sequences are obtained followed detection model is designed with the help of
by a convolutional neural network and then a Machine Learning (ML) algorithm so that
ROI is implemented. The YOLO algorithm motorcyclists can be identified [7]. Over the
is run in the next phase and it helps in past few years, there have been multiple
counting the vehicles [6]. The algorithm is solutions given to solve the problem of
trained based on the images and then the helmet detection. For example, it has been
object gets detected along with being observed that Support Vector Machines
tracked. The wrong way vehicle image is (SVM) has been made to use to classify
captured and an alert message is sends to helmets and human heads without helmets.
manage the traffic. This has been observed The edges of the helmet are computed based
to prevent traffic congestion and hence, on the possible helmet region and to do it,
traffic management takes place effectively. Canny Edge Detector has been applied. It
has been proposed by Waranusat to detect
moving objects with the help of a k-NN
classifier so that the head of the
motorcyclists can be detected. The
methodology that has been proposed
consists of the user, system, and admin as
the three most important entities.
Figure 1: Proposed system of Traffic
The system has been developed to identify
management using ML
the motorcyclists without helmets so that the
(Source:
traffic accidents can be managed which has
http://ijirt.org/master/publishedpaper/IJIRT1
been one of the major reasons of road
50994_PAPER.pdf)
congestion due to accidents taking place
from it. There is the inclusion of three
IV. Smart Traffic Management System
modules- user, system, and admin. In the
with help of Machine Learning
preliminary process techniques, there is the
Accidents of motorcycles have been on the
removal of the background noise and also
rise in recent years and the use of bicycles is
the contrast is enhanced along with
also on the rise. It has been necessary to
banalization of images. In the second stage,
detect the helmet as objects so that road
there is a classification of the vehicles by
safety is obtained. A customized object
taking in consideration aspect ratio and area surveillance. The data input that is given
as parameters. In the third stage, after consists of various speed limits of the roads
extraction of the head part classifier, which and there are sensors connected with the ITS
is given as input and the ML algorithm is [8]. The sensors take in the input of the
trained based on various images of helmets speed of the approaching vehicle and match
[7]. Through the help of trained features, it it with the speed limit. The data taken as
will take into consideration whether the input is matched with the help of the taken
motorcyclist is wearing the helmet or not. data in the database. If the speed limit
There is the generation of a database that crosses then the system sends the alert signal
consists of the time, date, and location of all [9]. Another key aspect of ITS is that of the
the bikes that are riding without the use of a data transmission which is done in real time
helmet. basis and the information is exchanged with
Application of Machine Learning help of various telecommunication units.
Algorithms for Intelligent Transportation Information can be communicated by the
System (ITS) announcement done through SMS, Internet
With the help of the Machine Learning and on-board units of vehicles.
Algorithm, it has been possible to detect the
traffic congestion and also forecast the V. The role of Machine learning in traffic
number of vehicles that will be present in prediction
the road. It has been observed that Machine The application of machine learning has
Learning algorithms are inherent elements of been majorly crucial in the transportation
the network traffic forecasting methodology. industry. The issue of traffic is significantly
The ITS Architecture has been explained in problematic for any country in the world.
this section. It has been observed that the Therefore, predicting the expected traffic in
first process that takes place is that of an area can benefit people and also reduce
automated data collection. This has been misuse of resources. Through the help of
done with the help of the Automatic vehicle Machine learning or ML computer systems,
identification process through the help of a traffic prediction can take place [5]. The
GPS-based vehicle locator and cameras. The algorithm of ML helps in learning
traffic count has been done with the help of relationships between the input variables and
a GPS-based locator and also it helps in the labels that help in predicting traffic
prediction. With the help of the The efficient prediction of broad traffic has
unsupervised learning opportunity provided the capability of creating an intelligent
by ML, some algorithms help in learning the transportation system. For analysing the
patterns on testing data and cluster the data level of traffic of a particular area, it is
for predicting future values. Hence, imperative to consider ML algorithms.
predicting the expected level of traffic Collecting information related to traffic,
becomes effective. Similarly, ML-driven such as location measured with GPS,
non-recurrent concept drifts are recognised direction, Start-End junction and speed need
and utilised for updated traffic propagation. to be analysed [3]. Calculation of these data
This also reflects the traffic flow prediction by calculating the distance between each
models in real-time [4]. The entire road vehicle within the specific junction provides
network is impacted by the development of a clear prediction. It is imperative to
the real-time algorithms of machine consider that for using ML in traffic
learning. prediction, creating an application for
providing GPS coordinates is crucial. On the
other hand, due to a lack of availability of
historical data, impact predictions of traffic
are often conducted using mathematical
models-based simulations [4]. There are
often considerations of suboptimal situations
as real-world traffic conditions can differ
from the considered assumptions.

Figure 2: Key elements of Machine

Learning
(Source: Zantalis et al., 2019)
 Predicting the traffic in a road network also
relates to identifying the congestion in a
particular area. If the level of congestion is
realised, it can be effective to manage the
roads and transportation system efficiently.
For controlling the traffic lights, visual
message system and lane signals and traffic
information, the application of machine
learning can be significantly efficient [1].
The Apache Hadoop has been utilised as the

Figure 3: Intelligent transportation machine learning method for predicting a

system highway travel time. This model has used

(Source: Meena, Sharma and Mahrishi, the highway toll collection data for analysis.

2020) From this model, it is realised that highway

drivers avoid traffic congestion and

The prediction of the traffic in a road minimise travel time by choosing departure

network depends on different aspects such times [10]. Similarly, in the case of arterial

as the general traffic level of that area, the road networks, surrounding arterial roads are

situation of the road. There are a significant often operated in the near saturation level.

number of traffic flow prediction models; Therefore, any non-recurrent incident, in this

however, they are failing due to beingunable case, results in significant congestion [4].

to process enormous datasets [3]. This Significantly heavy traffic congestion befalls

problem can be resolved with ML in the frequent time periods.

integration. The use of regression analysis is

helpful in prediction tasks of ML [5]. The
problem of classification of different kinds
of traffic can be resolved with the help of
ML also.

VI. Machine Learning and management

of traffic congestion
 flexibility for automatically controlling the
signal timers for minimising the congestion
in traffic. Therefore, soft real-time systems
of traffic management, time is vital. Hence,
data received by the present signal sensors in
a delayed manner can affect the solution
performance of MSR2C-ABPNN. Including
the artificial neural network time series can
be crucial in providing insights into the
traffic congestion at a particular road. On the
other hand, traffic prediction can be done
based on the deep neural network model
Figure 4: General structure of the
which is highly data-driven [4].
methods utilised to estimate and predict
traffic parameters
(Source: Alsrehin, Klaib and Magableh,
2019)

The application of ML in traffic

management is majorly in relation to traffic
congestion. The Smart Road Traffic
Congestion control with the application of
ML for road traffic can increase
transparency, efficiency and availability
offered to the citizens [2]. This neural
network is called MSR2C-ABPNN which
has helped in analysing the congestion Figure 5: MSR2C-ABPNN system model

frequency in that particular area. It is (Source: Ata et al., 2019)

imperative to consider that the impact of

smart transportation systems can be only The management of traffic congestion is

seen in smart cities. Furthermore, the only possible when there is adequate and

conventional systems have a lack of effective data that predicts the congestion. It
is relevant to consider that using ML congestion of the roads can be managed also
algorithms can provide insights into with the insights provided by the ML
managing traffic in road networks. However, system.
considering the time and place can affect the
prediction. ACKNOWLEDGEMENT
The research regarding ML-driven traffic

VII. Conclusion management has been completed

The current research paper has attempted to successfully. A varied range of knowledge

provide insights into ML-driven traffic has been gained by conducting this research.

management. From the findings of the study, I would like to thank my supervisor and

it is evident that there is a significant other faculties of the University who have

contribution of ML-driven traffic been significantly helpful in completing the

management. The traditional traffic research. I am also grateful to my peers. I

prediction models have been used over the would like to thank my family and friends

years to help drivers on highways and roads. for supporting me during the completion of

However, these conventional traffic the research.

prediction systems were often unsuccessful

in predicting traffic patterns. The traditional REFERENCES
systems of managing road traffic have often [1] Alsrehin, N.O., Klaib, A.F. and
failed to incorporate large sets of data for the Magableh, A., 2019. Intelligent
active application of the system [6]. transportation and control systems using
Therefore, consideration of the ML data mining and machine learning
integrated traffic management helps in techniques: A comprehensive study. IEEE
releasing the availability of vehicles. These Access, 7, pp.49830-49857.
systems can also contribute to transparent [2] Ata, A., Khan, M.A., Abbas, S., Ahmad,
results which help in handling congestion on G. and Fatima, A., 2019. Modelling smart
road networks. Therefore, the use of ML can road traffic congestion control system using
be beneficial for the road network. The machine learning techniques. Neural
findings of the study states that in applying Network World, 29(2), pp.99-110.
Machine learning in traffic congestion [3] Meena, G., Sharma, D. and Mahrishi,
control, GPS needs to be considered [3]. The M., 2020, February. Traffic Prediction for
 Intelligent Transportation System using Kharat/publication/327073557_Intelligent_
Machine Learning. In 2020 3rd International Road_Traffic_Control_System_for_Traffic_
Conference on Emerging Technologies in Congestion_A_Perspective/links/5bbd57542
Computer Engineering: Machine Learning 99bf1049b786b1b/Intelligent-Road-Traffic-
and Internet of Things (ICETCE) (pp. 145- Control-System-for-Traffic-Congestion-A-
148). IEEE. Perspective.pdf
[4] Nallaperuma, D., Nawaratne, R., [9] Wahab, L., & Jiang, H. (2019). A
Bandaragoda, T., Adikari, A., Nguyen, S., comparative study on machine learning
Kempitiya, T., De Silva, D., Alahakoon, D. based algorithms for prediction of
and Pothuhera, D., 2019. Online incremental motorcycle crash severity. PLoS one, 14(4),
machine learning platform for big data- e0214966.
driven smart traffic management. IEEE [10] Zadobrischi, E., Cosovanu, L. M.,
Transactions on Intelligent Transportation &Dimian, M. (2020). Traffic Flow Density
Systems, 20(12), pp.4679-4690. Model and Dynamic Traffic Congestion
[5] Zantalis, F., Koulouras, G., Karabetsos, Model Simulation Based on Practice Case
S. andKandris, D., 2019. A review of with Vehicle Network and System Traffic
machine learning and IoT in smart Intelligent Communication. Symmetry,
transportation. Future Internet, 11(4), p.94. 12(7), 1172.
[6] Ijirt.org, (2021), Traffic Management
System using Machine Learning Algorithm,
http://ijirt.org/master/publishedpaper/IJIRT1
50994_PAPER.pdf
[7] irjet.com, ( 2019), The Smart Traffic
Management System,
https://irjet.com/archives/V8/i4/IRJET-
V8I4732.pdf
[8] Researchgate.net, (2018), Intelligent
Road Traffic Control System for Traffic
Congestion:
A Perspective,
https://www.researchgate.net/profile/Vilas-

View publication stats