Traffic Signal Violation Detection Using
Traffic Signal Violation Detection Using
Volume 12, Issue 2, February 2021, pp. 207-217 Article ID: IJARET_12_02_019
Available online at http://iaeme.com/Home/issue/IJARET?Volume=12&Issue=2
ISSN Print: 0976-6480 and ISSN Online: 0976-6499
DOI: 10.34218/IJARET.12.2.2021.019
ABSTRACT
The number of new vehicles on the road is increasing rapidly, which in turn causes
highly congested roads and serving as a reason to break traffic rules by violating them.
This leads to a high number of road accidents. Traffic violation detection systems using
computer vision are a very efficient tool to reduce traffic violations by tracking and
Penalizing. The proposed system was implemented using YOLOV3 object detection for
traffic violation detections such as signal jump, vehicle speed, and the number of
vehicles. Further, the system is optimized in terms of accuracy. Using the Region of
interest and location of the vehicle in the duration of frames, determining signal jump.
This implementation obtained an accuracy of 97.67% for vehicle count detection and
an accuracy of 89.24% for speed violation detection.
Key words: Convolutional neural network, speed violation, signal jump, traffic
violation, YOLOV3
Cite this Article: S. Raj Anand, Naveen Kilari and D. Udaya Suriya Raj Kumar, Traffic
Signal Violation Detection using Artificial Intelligence and Deep Learning,
International Journal of Advanced Research in Engineering and Technology (IJARET),
12(2), 2021, pp. 207-217.
http://iaeme.com/Home/issue/IJARET?Volume=12&Issue=2
1. INTRODUCTION
Traffic violation detection systems are being developed as the number of vehicles on the
road is increasing at a large rate. This increase in number makes it difficult for trigger base
traffic violation detection system to keep up with the high volume of traffic and it is not
designed to detect multiple traffic violation at the same time for a given vehicle as a result
for an increase in traffic rule violation due to the lack of advance detection. The traffic
rule violation leads to various road accidents, traffic administration poses numerous basic
difficulties in most present-day urban areas. Manual checking of vehicles is troublesome,
and mistake-inclined due to feeble and problematic human memory. Consequently, a need
arises for a traffic violation detection system to deal with this errand, which can identify
criminal traffic offenses , for example, signal jump, over speeding and vehicle count. The
first traffic monitoring began with human traffic cops assigned at every junction to monitor
traffic, which required human resources and presence at all times. This consequently
became difficult with the increase in vehicles. This gave way to trigger based traffic
detection systems, which were specialized systems meant for detecting one type of
violation – speed. These were costly, could only be placed at one point on the road, and
were easy to avoid. A new system was required that could be operational 24x7, with least
or no human resource requirement and which could identify multiple violations with high
accuracy. This is how traffic violation detection using computer vision came into being.
Traffic violation detection using computer vision is mainly based on computer vision
technologies, which are related to image processing, artificial intelligence, and deep
learning. This detects objects with instances of semantic objects and class in digital images
and videos. The video captured from the camera is reused for various other domains and
hence the operation cost is significantly reduced. Monitoring infrastructure can be scaled
outside the city to rural &highways without additional cost. Since video captured from
moving vehicles, the whole city is covered instead of a few signals and junctions. This
also enables traffic police to capture traffic violations happening in small lanes to
highways. This does not have the one-time infrastructure cost involved in other monitoring
systems like Cameras. It saves time and relieves traffic police personnel to focus on other
productive work. Computer vision-based detection is based on the principle of vehicle
classification, environment awareness, and traffic violation detection.
Detection system, the proposed architecture of surveillance system with intelligent
detection and tracking of multiple vehicles from the surveillance input video using
YOLOv3 as an object detection algorithm. This is done through a neural network and an
object detection model which are used in the classification of the moving objects into
different respective classes, thus achieving vehicle classification. Next, from the same
given video footage, traffic lights, zebra crossing, different lanes, and traffic signs are
classified this comes under environment awareness. Combining these two, now violations
are detected based on violations are then detected these can occur on the road which are
signal jump, speed detection, and vehicle count. The main objective is to detect multiple
vehicle violation detection and it gives a more detailed picture of concepts and technology
involved in creating a traffic violation detection system using computer vision. It also aims
to throw light on some of the applications and the latest developments being made in the
said field.
Image Processing
Image processing refers to the analysis and manipulation of a digital image to enhance its
quality and interpret information present in it.
YOLOv3
YOLOv3 algorithm is mainly constructed of a particular connected Convolutional Neural
Network. Layers of CNN is a class of deep neural networks (DNN) to analyze digital
images or videos. It is a class of multi-layer perceptron, refers to a fully connected network
(FCN) in which all neurons of one layer are connected to other neurons in the next layer.
CNN is a specialized category of neural networks used effectively to process data in a grid-
like manner. Parameter sharing is a distinguishing feature of CNNs, which is of great
significance in object detection. CNN's are different from fully connected forward neural
networks, each of its trained weight is used, such that in overall CNN architecture every
component makes use of the kernel in the input position, meaning that in each location and
position, a category of all the parameters is learned instead of a different category of
parameters. This property is very valuable in encapsulating the whole entire environment.
Figure 2 shows the bounding box prediction. CNN takes input as a digital image and
output is
• Objects with boundary boxes, Coordinates of this boxes and positions concerning
the frame,
• The probability score of all the object in a digital image in the respective bounding
box’s,
• Probabilities on every object present inside the frame in its bounding box.
The detection of the object is found in three different layers, where input dimensions
of the given digital image are in height and width, those are 13x13, 26x26, and 52x52.
Detection takes place in three various scales. The output of the digital image from the three
deleted output layers have the same height and width concerning the inputs digital image,
but depth will always depend on the definition:
Depth = (4 + 1 + class probabilities) х 3,
where, four denotes properties of bounding boxes such as height (bh), width (bw), x
and y denote the position of the given bounding boxes (bx, by) inside a given digital image,
one is the probability of the detected object in the box (pc) and class probabilities are the
probabilities for each class (c1, c2, …). That sum is multiplied by three, this done
Cause each of the cells predicts 3 bounding boxes in a grid. This results in the output
of 10647 bounding box predictions in every digital image.
3.2. Detector
Once the three features vectors are obtained, it can be fed into the detector. Through this
config file, multiple 1x1 and 3x3 Conv layers are used before a final 1x1 Conv layer to
form the final output. For medium and small scale, it also concatenates features from the
previous scale. By doing so, small scale detection can also benefit from the result of large-
scale detection [7].
Figure 11. Detection of a traffic violation output frame: bounding box and violation detection
(signal jump)
Figure 13. Failure to the detection of a traffic violation Figure 14. Output Parameters of entering the frame
Figure 14 shows that the car is not at the ROI, which is the traffic signal line, hence the
car id not detected as violating the traffic rules. Since this model is capable of detecting
vehicles in every frame it is having been enhanced to detect the speed of the vehicle in a
video. The accuracy of detecting the speed of the vehicle was 5.4% times slower than the
actual speed. The corresponding frame of the video is shown in Fig 9, the shown frame
detects the vehicles and also the respective speed of the vehicle. This is done using the
pixel calculation, area of the boundary box, and the anchor points.
Analysis of vehicle detection and its time tabulated as shown in table I and II.
Veracity
Computational power is required in both hardware and software, together with a large
amount of computing time, to identify objects, labels, and their respective location in an
image or video frame. Under partial concealment, resolution of the interested detection
must improve localization accuracy on small objects. A centralized anomaly detection
system is not realistic and feasible for real-time analysis of data.
6. CONCLUSIONAND FUTURESCOPE
6.1. Conclusion
Detections of traffic violation in the video surveillance is challenging as the number of
vehicles on the road and traffic rules are depended on the different area of the road and
timings. This paper proposes that the YOLOv3 algorithm is suitable for traffic violation
detection. Results show that the detection of multiple traffic violations from a single input
source is achievable. The system has an accuracy of 97.67% for vehicle count detection
and an accuracy of 89.24% to detect the vehicle speed. The detection time is lower for
high dense traffic flow. Thus, the system operation speed is dependent on the density of
traffic.
Further, it can be implemented for larger datasets by training using GPUs and high-end
FPGA kits.
REFERENCES
[1] X. He et al., “A Driving Warning Method based on YOLOV3 and Neural Network”, IEEE
International Conference on Service Operations and Logistics, and Informatics(SOLI),
Zhengzhou, China, 2019.
[2] H. Qu et al., “A Pedestrian Detection Method Based on YOLOv3 Model and Image Enhanced
by Retinex,” 11th International Congress on Image and Signal Processing, Bio Medical
Engineering and Informatics(CISP-BMEI),Beijing,China,2018.
[3] J.Wonetal.,“AnImprovedYOLOv3-basedNeural Network for Deidentification Technology,”
34th International Technical Conference on Circuits/Systems, Computers and
Communications (ITC-CSCC), Korea,2019.
[4] P. T umaset al., “Automated Image Annotation based on YOLOv3”, IEEE 6th Workshop on
Advances in Information, Electronic and Electrical Engineering (AIEEE),Vilnius,2018.
[5] Krishnaetal.,“Automated traffic monitoring system using computer vision”2016.
[6] P. Manochaet al., “Korean Traffic Sign Detection Using Deep Learning”, International SoC
Design Conference (ISOCC), Korea, 2018.
[7] Y. Lee et al., “Fast Detection of Objects Using a YOLOv3 Network for a Vending Machine,
”IEEE International Conference on Artificial Intelligence Circuits and
Systems(AICAS),Taiwan,2019.
[8] J. Wangetal., “ Real-Time Non-Motor Vehicle Violation Detection in Traffic Scenes”, IEEE
International Conference on Industrial Cyber Physical Systems(ICPS),Taipei,Taiwan,2019.
[9] A. Corovicet al., “The Real-Time Detection of Traffic Participants Using YOLO Algorithm,”
26th Telecommunications Forum (TELFOR), Belgrade,2018.
[10] N. Krittayanawachet al., “ Robust Compression Technique for YOLOv3 on Real-Time Vehicle
Detection”, 11th International Conference on Information Technology and Electrical
Engineering (ICITEE),Pattaya,Thailand,2019.
[11] github.com/anmspro/Traffic Signal Violation-Detection-System.
[12] Mohanaet.al., “Simulation of Object Detection Algorithms for Video Survillance
Applications”, 2 ndInternational Conference on I-SMAC (IoT in Social, Mobile ,Analytics and
Cloud),2018.
[13] YojanChitkaraet. al.,“ Background Modelling techniques for foreground detection and
Tracking using Gaussian Mixture model” International Conference on Computing
Methodologies and Communication,2019.
[14] Mohanaet.al., “Performance Evaluation of Background Modeling Methods for Object
Detection and Tracking,” International ConferenceonInventiveSystemsandControl,2020.
[15] N. Jain et.al.,“Performance Analysis of Object Detection and Tracking Algorithms for
Traffic Surveillance Applications using Neural Networks,” 2019 Third International
conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud),2019.
[16] Ruben J Franklin et.al.,“Anomaly Detection in Videos for Video Surveillance Applications
Using Neural Networks,” International ConferenceonInventiveSystemsandControl,2020.
[17] Abhiraj Biswas et. al., “ Classification of Objects in Video Records using Neural Network
Framework,” International conference on SmartSystemsandInventiveTechnology,2018.