Abstract
The Traffic Signal Violation Detection System is an innovative solution leveraging computer vision techniques to enhance traffic management and improve road safety. This project aims to develop an intelligent system capable of automatically detecting and monitoring traffic signal violations at intersections. Using computer vision algorithms, the system processes video feeds from surveillance cameras installed at traffic signals. The proposed approach involves multiple steps, including video preprocessing, object detection, and violation classification. In the preprocessing phase, the video frames are analyzed to extract relevant information. This step helps in enhancing the overall accuracy of subsequent stages. The system employs state-of-the-art object detection algorithms such as Ultralytics YOLOv8 (You Only Look Once Version 8) and SORT (Simple online real-time tracking) to detect vehicles within the video frames. By accurately identifying objects, the system can analyze their behavior and adherence to traffic rules. The classification phase involves determining whether a vehicle has violated the traffic signal rule i.e. red light running To achieve robust performance, the project utilizes a comprehensive dataset of labeled traffic scenarios to train and fine-tune the deep learning models. We have used COCO (Common Objects in Context) dataset which contains more than 348K images. We have also trained a custom model to detect red, yellow and green light and pedestrian crossing.