Traffic Signal Violation Detection System using YOLOv3

The quantity of vehicles on the road is progressively escalating, resulting in heavily congested roads. This could result in many accidents. Computer vision systems for spotting traffic violations are an effective way to cut down on rule-breaking by monitoring and giving warnings. The proposed system is to catch instances of breaking traffic rules, like jumping signals or crossing the center line. This work utilizes You Only Look Once (YOLOv3), renowned for its ability to process data in real-time, to boost the precision of detection. Furthermore, the system undergoes optimization for improved accuracy. Additionally, the Convolutional Neural Network (CNN) model has been implemented to make the recognition of small violations even better. The system looks closely at where the vehicle is in each frame, especially focusing on catching instances of jumping signals, using a Region of Interest (ROI) method. This work focuses on the efficient alerting mechanism within a traffic rule violation detection system, ensuring timely notifications to relevant authorities. With the implementation of this detection system, the positive outcomes on public safety are delved into, underscoring the ability to mitigate the risk of accidents. This implementation achieved a vehicle count detection accuracy of 98.76%.