Intuitive Traffic Regulation Breach Detection using Deep Learning

Intuitive Traffic Regulation Breach Detection using Deep Learning

Mrs. Jayandhi.G

Assistant Professor

Electronics and Communication Engineering Velammal Engineering College

Chennai 600066

Jaya Chithra P

Electronics and Communication Engineering Velammal Engineering College

Chennai 600066

Kruthika S

Electronics and Communication Engineering Velammal Engineering College

Chennai 600066

Mansi Milan Dave

Electronics and Communication Engineering Velammal Engineering College

Chennai 600066

Rajasri K

Electronics and Communication Engineering Velammal Engineering College

Chennai 600066

AbstractThe majority of vehicles on roads in India are increasing faster because of which traffic management has become one of the main problems. When every traffic infraction is frequently found, traffic may be managed effectively. Because it is ineffective to detect traffic rule violations using conventional/manual methods and current technologies, traffic management has become highly challenging. In this research, a system that uses image processing to identify serious infractions including excessive speeding and helmet identification along with a number plate recognition method that will simplify traffic control is presented. For number plate recognition and extraction as well as tracking and violation detection, we advise using Yolo-v4 Tesseract and Yolo-v4 DeepSORT, respectively. The created system will be especially helpful in evaluating various security legislation, assisting in the enforcement of stringent traffic laws, and advancing the creation of smart cities.

Keywords Artificial intelligence (AI), traffic infraction, smart city.

1. INTRODUCTION

   In the current, evolving world, traffic law infractions are becoming a major issue for the majority of rising nations. Traffic offenses and the amount of motorcycles upon that roadway are both dramatically rising. It has always been challenging and dangerous to detect traffic offenses. Despite automation in traffic management, it remains a very challenging task. Variable rotations, different plate sizes, and inconsistent lighting when an image was captured. This project's main goal is to effectively and appropriately manage traffic regulation breaches. The suggested model incorporates an automated system for image capturing that uses a computer and a camera. The project offers techniques as well as further image- manipulation to identify characters and locate plates in order to detect number plates more rapidly and easily. This project's implementation of number plate recognition stops there. A camera on this device can capture a photo, find a number in the

   image, and then use a character recognition programme to retrieve characters from the picture. In order to avoid accidents, strict rules are required because motorcycles are inexpensive and widely used. Because it is necessary by law to wear a helmet, breaking traffic rules entails severe penalties. methods. The automated end-to-end system is for detecting infractions and issuing tickets. The system is made up of three main parts: vehicle tracking, vehicle detection and number plate recognition.

2. COMPONENTS AND ALGORITHMS USED

   1. RASPBERRY PI 4

      Deeply embedded apps can use the Raspberry Pi 4's capability in a small form factor. A quad-core ARM Cortex-A72 CPU, two footage outputs, number of additional connections are also included and the Raspberry Pi Compute Module 4 has many other features. With a selection of RAM and eMMC Flash choices, as well as with or without wifi access, and available in 32 versions. The suggested operating system for regular use of a Raspberry Pi is Raspberry Pi OS, formerly referred as Raspbian. Using the Raspberry Pi Imager, you can quickly and easily load the Raspberry Pi OS and other operating systems to a microSD card so that they are available to use with your Raspberry Pi. Over 35,000 packages, or precompiled programmes, are included with the System, making it simple to run on your Raspberry Pi. Raspberry Pi OS is actively being developed, with an emphasis on improving the dependability and usefulness of Debian packages as is practical on Raspberry Pi. The Raspberry Pi 4 is powered by the Broadcom BCM2711 Chipset, contains a quad-core, 64-bit ARM Cortex-A72 CPU operating at 1.5 GHz (later models: 1.8 GHz) and a shared L2 cache of 1 MB. Unlike earlier versions, which were all equipped with specialized interrupt drivers Although the interrupt driver on this SoC is incompatible with virtualization, it is interoperable with the When using ARM virtualization capabilities, interrupt spread is supported by hardware thanks

      to the ARM Generic Interrupt Controller (GIC) design 2.0. The quad-core Cortex-A72 Processor in the Raspberry Pi 4 is said to be three times as fast as that in the Raspberry Pi 3. The Raspberry Pi 4 is available with one, two, four, or eight gigabytes of Memory. The 1 GB model had been available at the time of the device's launch in June 2019 but had been phased out by March 2020. To replace it, the 8 GB version was released in May 2020. The 1 GB version was made available again in October 2021. .

      Fig 1 : Raspberry Pi 4

      While being usually pre-configured to operate as a headless computer, the Raspberry Pi may be used with any common USB computer keypad and mouse. It can also be utilized with USB storage, USB to MIDI converters, and nearly any other component or device having USB capabilities, according to the loaded device drivers in the base operating system (many of which are included by default). The surface of the Raspberry Pi has a number of pins and contacts that enable the installation of extra hardware.

   2. YOLOV4

      For the MS COCO dataset, a cutting-edge real-time object identification system named YOLO, which stands for "You Only Look Once," provides 43.5% Average Precision (65.7% AP50) at a real-time speed of roughly 65 FPS on a Tesla V100 (for yolov4). In addition to hint-generating suggestion systems, the main objective of YOLO is to improve real-time object identification accuracy, enabling them to be used for stand- alone process management and minimizing human input. One- staged and two-staged object detectors are two separate versions. Two-stage detectors work in two steps: first, significant regions are recognised, and areas are classified to see if the object is identified in that particular area. In terms of accuracy and speed, R-CNN and Fast R-CNN are surpassed by the single staged object detector YOLOv4, as are two staged detectors.

      Fig 2: Structure of YOLO V4

      Backbone

      YOLOv4 essentially bases its architecture on one of the three types. The CSPResNext50 and CSPDarknet53 models are two of the three feature extractors.

      Neck

      In this stage, characteristics that were previously formed in the backbone are collected in preparation for subsequently feeding them to the head for detection.

      Head

      In YOLOv4, the primary goal is to conduct prediction, which involves classifying data and regressing bounding box boundaries. The YOLOv3 head is employed. It provides details about the bounding box coordinates (x, y, h, w). It comprises the label's width, height, center, and prediction score. Each anchor box can be equipped with a YOLOv4 head.

      C.CNN

      CNN is a potent algorithm for processing images. The best algorithms available right now for automating picture processing are these ones. For tasks like identifying the objects in an image, many businesses employ these algorithms. Images are composed of RG data. Image import from a file into memory is possible with Matplotlib. The computer does not see any visuals; it only sees a collection of numbers. Color images are stored in arrays that are three dimensional. The first two dimensions represent the height and breadth of the image (the number of pixels). The red, green, and blue hues of each pixel are depicted in the final dimension.

      Fig 3: Structure of CNN

      Image and video recognition applications use convolutional neural networks. CNN is typically used as segmentation, object

      classification, and image recognition functions in programs for image analysis. The layers in convolutional neural networks can be divided into three categories.

      1. Convolutional Layer: In a typical neural network, the convolutional layer connects each input neuron with the subsequent convolution layer. In CNN, the input layer neurons only sometimes interact with the neuron convolution layer..

      2. Pooling Layer: The pooling layer reduces the feature's level of dimensionality map. There will be several triggering and pooling layers inside the CNN's convolution layer.

      3. Fully-Connected layer: These layers refer to the uppermost tiers of network. After getting flattened, the output from the last pooling or convolutional layer is transferred into the fully connected layer.This is the layer's input.

   D.OCR

   Tesseract is an Apache 2.0-compliant OCR engine which is free and open-source. For programmers, it can be applied directly or through the use of an API to extract written text from photos. Many different languages are supported. Tesseract lacks a native graphical user interface, but there are many available on the 3rdParty website. By the use of wrappers, which may be obtained here, Tesseract works with a wide variety of frameworks and coding languages. It can be used along with the currently accessible layout analysis to locate text inside a large document, or it can be used along with a third- party text detector to locate information with an imagery of a plaintext line.

   Fig 4: OCR Process

3. PROPOSED MODEL

   Using a Convolutional neural networks module, we acquire the frames from the live stream as a source to execute violation recognition. In the Indian context, the offenses included for our analysis include not having a helmet on, using a cellphone during driving, multiple riding, wheeling, and not stopping. Each image frame is subjected to object detection using the YOLO-v4 intrusion detection system, which can identify one or more infractions.

   1. Object detecting model: To prevent tedious counting and to guarantee that the vehicles are captured across numerous frames, the identified object is then monitored using the vehicle detection model.

      Fig 5:Traffic violation detection

   2. License plate recognition: Each offending vehicle is then trimmed out using the dimensions deduced from the current frame provided by the object detection module after violations have been found and tracked. Using cropped car photos, YOLO-V4 is utilized to identify a vehicle's license plate. Tesseract is used to get the registration numbers from the number plate and save these in a library using OCR which stands for optical character recognition.

      Fig 6: Number plate detection

   3. Ticketing: Users of vehicles are informed of related offenses via the database. The database can also be utilized to do statistical research on traffic law infractions.

   A. Block Diagram

4. CONCLUSION

   The performance is mainly focused on Accuracy. Machine learning techniques provide an effective mechanism in detection and data mining as it overcomes the problem with traditional techniques. These techniques enhance the data optimization along with improving the efficiency with better results and greater predictability.

5. FUTURE SCOPE

   The system can be strengthened in the future by adding other categories of motorized vehicles and identifying matching vehicular traffic infractions. Furthermore, structures can be created to deal with many camera viewpoints and more different types of violations. The OCR for detecting license plates may additionally be trained to recognise number plates with unusual fonts and patterns. Moreover, experiments using more recent object identification models can be used to compare models' precision and rapidity.

6. REFERENCES

[1] Aniruddha Tonge, S. Chandak, R. Khiste, U. Khan and L. A. Bewoor, "Traffic Rules Violation Detection using Deep Learning," 2020 4th International Conference on Electronics, Communication and Aerospace Technology (ICECA), 2020, pp. 1250-1257,doi: 10.1109/ICECA49313.2020.9297495.

[2] Ruben.J Franklin and Mohana, Traffic Signal Violation Detection using Artificial Intelligence and Deep Learning, 2020 5 th International Conference on Communication and Electronics Systems (ICCES), 2020, PP. 839-844, doi: 10.1109/ICCES48766.2020.9137873.

[3] Chetan Kumar B, R. Punitha and Mohana, "Performance Analysis of Object Detection Algorithm for Intelligent Traffic Surveillance System," 2020 Second International Conference on Inventive Research in Computing Applications (ICIRCA), 2020, pp. 573,579,doi:10.1109/ICIRCA48905.2020.9182793.

[4] Siddharth Tripathi, Uthsav Shetty, Asif Hasnain, Rohini Hallikar,"Cloud Based Intelligent Traffic System to Implement Traffic Rules Violation Detection and Accident Detection Units", Proceedings of the Third International Conference on Trends in Electronics and Informatics (ICOEI 2019) IEEE Xplore Part Number: CFP19J32-ART; ISBN: 978- 1- 5386-9439- 8.

[5] Helen Rose Mampilayil and R. K., "Deep learning-based Detection of One-Way Traffic Rule Violation of Three- Wheeler Vehicles," 2019 International Conference on Intelligent Computing and Control Systems (ICCS), 2019, pp. 1453- 1457, doi: 10.1109/ICCS45141.2019.9065638.

[6] Ali entas, S. Kul and A. Sayar, "Real-Time Traffic Rules Infringing Determination Over the Video Stream: Wrong Way and Clearway Violation Detection," 2019 International Artificial Intelligence and Data Processing Symposium (IDAP), 2019, pp. 1- 4, doi: 10.1109/IDAP.2019.8875889.

[7] M. Purohit and A. R. Yadav, "Comparison of feature extraction techniques to recognize traffic rule violations using low processing embedded system," 2018 5thInternational Conference on Signal Processing and Integrated Networks (SPIN), 2018,pp. 154-158, doi: 10.1109/SPIN.2018.8474067

[8] S. P. Mani Raj, B. Rupa, P. S. Sravanthi and G. K. Sushma, "Smart and Digitalized Traffic Rules Monitoring System,"2018 3rd International Conference on Communication and Electronics Systems (ICCES), 2018, pp. 969-973, doi: 10.1109/CESYS.2018.8724086.

[9] Shashank Singh Yadav, V. Vijayakumar and J. Athanesious, "Detection of Anomalies in Traffic Scene Surveillance,"2018 Tenth International Conference on Advanced Computing (ICoAC), 2018, pp. 286-291, doi: 10.1109/ICoAC44903.2018.8939111.

[10] R. Shreyas, B.V.P.Kumar, H.B. Adhithya, B. Padmaja, "Dynamic traffic rule regulation monitoring system using automatic number plate recognition", 2017 2nd International Conference on Telecommunication and Networks (TEL – NET), Volume 1 , 2017, pp . 1-5 , doi : 10.1109/TEL- NET.2017.834352