Volume 8, Issue 3, March – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Smart Traffic Management using Deep Learning

B. B. Waghmode, Namokar Magdum, Vardhaman Patil, Vishal Kumar
Department ofComputer Engineering
Sinhgad Institute of Technology and Science, Nahre
Savitribai Phule Pune University

Abstract:- The recent era is marked by rapid Due to increase in the no. of roads and vehicles, traffic
improvement and advances in technology. One of the control has become an essential part of intelligent transport
most essential areas that demand improvement is the system. Many researches have been conducted for traffic
traffic signal, as it constitutes the core of the traffic management applications based on image and video
system. This demand becomes stringent with the processing approaches.[7] The analysis of traffic video data
development of Smart Cities. Unfortunately, road traffic includes detection of vehicles, counting of vehicles,
is currently controlled by very old traffic signals congestion of traffic and collisions of vehicles. These
regardless of the relentless effort devoted to developing applications have become popular recently due to the
& improving the traffic flow. These traditional traffic availability of low-cost cameras and embedded devices.
signals have many problems including inefficient time Real-time videos pose several challenges for automated
management in road intersections; they are not immune traffic analysis.[2] The difficulties faced by the automated
to some environmental conditions, like rain; and they system include the presence of shadows, occlusion of
have no means of giving priority to emergency vehicles. vehicles, Environmental variations such as rain, fog, dust,
In this paper, we present the architecture of our etc., which normally degrade the performance.
proposed Smart Traffic Signal controller. We present
local traffic management of an intersection based on the Existing traffic prediction approaches can be roughly
demands of future Smart Cities for fairness, reducing categorized into 2 types, that is, model-based approaches
commute time, providing reasonable traffic flow, and data-driven approaches.[4] A model-based approach
reducing traffic congestion. Traffic problem are mimics the traffic dynamics by explicitly adopting traffic
increasing day by day and it is becoming a serious flow models. Once the traffic simulation system is well
problem in the recent world [2]. Various traffic established, traffic prediction can be achieved through the
monitoring systems have been developed. Traffic is evolution process of the system. In comparison, a data-
always a complex and challenging problem due to a driven approach establishes a mapping between historical
mixture of different types of vehicles as well as the large traffic flow and its future prediction but does not presume a
number of vehicles on road. To improve the traffic traffic flow model to describe the dynamics of a
management, it is critical to develop a real time traffic transportation system. With the development of big data and
flow estimation system which can detect, classify and deep learning, data-driven prediction models have attracted
count vehicles, detect traffic violation at any given time. more attention and achieved superior performances.
In this study, a multi-vehicle detection and tracking However, as far as we can tell, there is no one traffic
approach was proposed to achieve these requirements. prediction model that consistently outperforms other models
under any traffic condition and for any traffic dataset. The
I. INTRODUCTION solution must be intelligent and has decision-making
capabilities depending on the different situations in the
The traffic congestion problem is increasing day by different lanes. There are different phases of the proposed
day everywhere around us and must be solved more model. The first phase is video processing. The continuous
intelligently considering the continuous increase in traffic in video stream from the traffic camera situated at the traffic
not-so distant future. Among most proposed solutions like signal junction is fed to the model. During the video
building more infrastructure or improving citizen commute processing, the frames are extracted from the video input.
patterns, creating an intelligent control system to manage the There are hundreds of frames present in a stream of few
traffic flow comes to be a better and most efficient solution. seconds, so it is not possible to process each frame and it is
The more intelligent control system part over here means to not required because there is no such significant change in
use more sophisticated timing algorithms with the help of the vehicle density between two consecutive frames.
machine learning algorithms to solve the problem Therefore, each frame after a particular time interval is
autonomously. Instead of using fixed time intervals system taken. Then shrinking the size of the frame is done from the
should be able to set variable timers to each lane.[1] These actual size of the frame to 416 x 416 pixels. After this phase,
variable timers should be computed by the algorithms based the processed frame is fed to the model for object detection.
on different traffic conditions in real-time. Using such The neural network extracts the features and identifies the
solutions in real life can positively affect various object using bounding boxes. The total vehicle count of a
characteristics of a daily commute, especially in lane is passed to the dynamic traffic signal timer algorithm.
metropolitan cities. It may also help in saving many valuable The algorithm considers the density of all the other lanes
resources like time and fuel while managing the smooth and calculates the relativity between them. Depending on
traffic flow with the least stops possible. which, it classifies the lane in either low, medium, or high

IJISRT23MAR1798 www.ijisrt.com 2136

Volume 8, Issue 3, March – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
vehicle density class, it decides the green signal timer for a [4] et al. proposed a solution that is efficient and fast which
particular lane. uses image processing, data mining, and artificial neural
network. The proposed method measures the average speed
II. LITERATURE REVIEW of traffic and traffic density from cameras in real time. The
processing speed was faster as well. However, only
Surging traffic levels and increasingly busier roads are prediction and estimation of the traffic system were done.
common issues across the globe. Consequently, there is an There is no proper solution to the traffic congestion
increasing requirement to develop intelligent traffic problem. To give an optimal solution regarding the traffic
surveillance systems that can play a crucial role in highway congestion, Jingwei [10] Cao et al. used a You Only Look
monitoring and road management systems. Author Jose Once and SSD (Single Shot multi-box Detector) model
Melo et al. have addressed challenges of lane detection with which are one of the current mainstream object detection
the help of segmentation and clustering methods. The frameworks based on deep learning. A Faster R-CNN and
algorithm they have used is the Public Transportation regression idea of YOLO realizes the detection and
facilities and Equipment Management System (PTMS) classification of multiple bounding boxes based on a simple
algorithm, which helps in resulting input to higher-level end-to-end network that works on the KITTI dataset. We
traffic monitoring systems like estimating traffic speed, additionally implemented the program of wrong side vehicle
frequency of lane changes, and accident detection. Alvin detection in addition to impose the fine for that vehicle.
Abdagic [5]et al. proposed a solution to memorize stationary
vehicles, detecting turning movement using an optical flow III. RELEVANCE
algorithm. Instead of relying heavily on the sensors, their
solution uses CPU processing. Thus, reducing the costs of To develop a system of smart traffic management
both development and maintenance. The process contains a using Deep Learning due to limited resources provided by
video capture using infrared (IR) cameras converted into current infrastructure and increase in number of road users.
frames of intensities of traffic. Using these intensities optical In various countries the traffic problems are being increased
flow is calculated and a magnitude is generated. The main day by day and traffic is becoming a more severe issue. The
problem which arises while detecting the vehicles is main cause for increment in traffic deadlocks is the
identifying the lane and vehicles present in that lane. In the increment in the number of transportation objects, the
proposed study, a system to count the number of vehicles on deprived development of roads and no appropriate allocation
roads is discussed. The method involves analyzing a of resources. To control the traffic is a very difficult job and
sequence of road images to determine the flow of vehicles only the traditional system cannot control this serious
for the given time and place. system. Traffic survey is the study to determine the number,
of movement, & classification of vehicles at a given time.
The proposed Surging traffic levels and increasingly The survey exercises are important for the relevant
busier roads are common issues across the globe. authorities to perform planning & design of traffic facilities,
Consequently, there is an increasing requirement to develop estimating road usage and traffic trends, measuring current
intelligent traffic surveillance systems that can play a crucial demand to decide priorities for improvement and road
role in highway monitoring and road management systems. expansion.
Author Jose Melo et al. have addressed challenges of lane
detection with the help of segmentation and clustering IV. MOTIVATION
methods. The algorithm they have used is the Public
Transportation facilities and Equipment Management The primary motivation for this kind of solution can be
System algorithm, which helps in resulting input to higher- said to be the increasing transportation in major cities. Fast
level traffic monitoring systems like estimating traffic speed, transportation systems and rapid transit systems are vital for
frequency of lane changes, and accident detection. Alvin [5] economic developments for any state. More population
et al. proposed a solution to memorize stationary vehicles, means more vehicles on the streets day-by-day. Traditional
detecting turning movement using an optical flow algorithm. traffic management systems are not built with the
Instead of relying heavily on the sensors, their solution uses consideration of increasing traffic at this high pace.[6] Thus,
CPU processing. Thus, reducing the costs of both it cannot keep up with the variable changing traffic
development and maintenance. The process contains a video conditions daily. In the past few years, fields like machine
capture using infrared (IR) cameras converted into frames of learning and computer vision have developed to be so better
intensities of traffic. Using these intensities optical flow is and easy to implement these days. Using concepts like deep
calculated and a magnitude is generated. The main problem learning to detect the vehicles and automatically adjust
which arises while detecting the vehicles is identifying the accordingly helps in many ways for better transportation
lane and vehicles present in that lane. A solution to this flow. An algorithm can be created to detect the patterns of
problem is given by Prashant Jadhav et al. using image the traffic flow and react accordingly. This not only makes
processing in MATLAB. [2] In the proposed study, a system the system robust but also more efficient in managing the
to count the number of vehicles on roads is discussed. The traffic. Traffic flow prediction is an essential part of the
method involves analyzing a sequence of road images to intelligent transport system. This is the accurate estimation
determine the flow of vehicles for the given time and place. of traffic flow in a given region at a particular interval of
The proposed method used an ANN model to estimate time in the future. The study of traffic forecasting is useful
vehicle densities, traffic flow rates, and vehicle trajectories in mitigating congestion and make safer and cost-efficient
on different video captures of road intersections. Surojit Dey travel. Traffic management is the arranging, observing and

IJISRT23MAR1798 www.ijisrt.com 2137

Volume 8, Issue 3, March – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
control or impacting of activity. It expects to: boost the traditional static timer. The main advantage of the proposed
adequacy of the utilization of existing foundation; to method is the use of video processing over sensors which
guarantee dependable and safe operation of transport; to reduces the setup cost, low maintenance, and relatively more
address ecological objectives; and guarantee reasonable durability and accuracy.
assignment of framework space among contending clients.
Traffic monitoring system developed so far are primarily REFERENCES
focused on structured traffic that is not the case in a country
like ours. Development of overhead structures can’t be [1]. Indrabayu , Rizki Yusliana Bakti , Intan Sari Areni , A.
considered as avail an option since it increases the cost Ais Prayogi “Vehicle Detection and Tracking using
substantially, the same goes for under the road construction. Gaussian Mixture Model and Kalman Filter” , 2016
Necessity to analyze traffic pattern, near real time reporting International Conference on Computational
and simultaneous conduction of smooth traffic-flow. Our Intelligence and Cybernetics Makassar , Indonesia , 22-
country (India) is the second largest population of world; 24 November 2016
according to that vehicles are increased day to day life. [2]. Li Xun , Nan Kaikai , Liu Yao , Zuo Tao “A Real-
Here, the questions arise! how to avoid the congestion in the Time Traffic Detection Method Based on Improved
road; that means traffic management. [Google] Kalman Filter”, 2018 3rd International Conference on
Robotics and Automation Engineering (ICRAE)
Guangzhou, China , 17-19 November 2018
[3]. Safoora Maqbool, Mehwish Khan, Jawaria -Tahir,
Abdul Jalil, Ahmad A li, Javed Ahmad Vehicle
Detection, Tracking and Counting” 2018 IEEE 3rd
International Conference on Signal and Image
Processing (ICSIP) Shenzhen, China, 13-15 July 2018
[4]. R. Krishnamoorthy, Sethu Manickam “Automated
Traffic Monitoring Using Image Vision”. The 2nd
International Conference on Inventive Communication
and Computational Technologies (ICICCT 2018)
Coimbatore, India, 20-21April 2018
[5]. Jess Tyron G. Nodado, Hans Christian P. Morales, Ma
Angelica P. Abugan , Jerick L. Olisea , Angelo C.
Sample Dataset 1 Aralar , Pocholo James M. Loresco “ Intelligent Traffic
Light System Using Computer Vision with Android
Monitoring and Control” , Proceedings of TENCON
2018 - 2018 IEEE Region 10 Conference Jeju
[6]. Sayan Mondal, Alan Yessenbayev, Jahya Burke, Nihar
Wahal, “A Survey of Information Acquisition in
Neural Object Detection Systems”, 32nd Conference
on Neural Information Processing Systems (NeurlPS
2018), Montreal, Canada
[7]. Muhammad Hanif Tunio, Noor Ahmed Shaikh, Imran
Memon, Riaz Ahmed Shaikh, Ghulam Ali Mallah,
Yumna Magsi Automation of Traffic Control System
Using Image Morphological Operations” 2020
International Conference on Information Science and
Communication Technology 978-1-7281-6899-
9/20//$31.00 ©2020 IEEE
Sample Dataset 2 [8]. Asha C S, A V Narasimhadhan “Vehicle Counting for
Traffic Management System” 978-1-5386-1112-
V. CONCLUSIONS 8/18/$31.00 ©2018 IEEE
[9]. Mingqi Lv , Zhaoxiong Hong, Ling Chen , Tieming
In this article, a smart traffic control system is Chen, Tiantian Zhu , and Shouling Ji , Member, IEEE
proposed which can be widely used in a smart city “Temporal Multi-Graph Convolutional Network for
application for optimal traffic flow of vehicles. For smooth Traffic Flow Prediction” 1524-9050 © 2020 IEEE.
flow of traffic, it is necessary to handle traffic congestion https://www.ieee.org/publications/rights/index.html
properly. The system takes the real-time traffic data, [10]. Dongfang Ma, Xiang Song, and Pu Li “Daily Traffic
captured from the pre-installed cameras at the intersection as Flow Forecasting Through a Contextual Convolutional
input for the object detection phase and calculates the Recurrent Neural Network Modeling Inter-and Intra-
number of vehicles for density estimation. These densities Day Traffic Patterns” 1524-9050 © 2020 IEEE.
are given to the timer algorithm as input for manipulating https://www.ieee.org/publications/rights/index.html
the green signal timer of the consecutive lanes. The timer
algorithms are capable of handling multiple scenarios based
on the practical condition which removes the flaws of the

IJISRT23MAR1798 www.ijisrt.com 2138