Traffic Rule Violation Detection using Detectron2 and Yolov7

With the absence of a reliable public transportation system, motorcycles have gained popularity as a means of commuting. However, this surge in motorcycle usage has brought about a concerning increase in accidents. The major contributing factors to these accidents include faulty licensing laws, inadequate road conditions, a disregard for helmet usage, wrong lane violations, and the practice of triple riding. Our system is designed to identify and enforce penalties for various violations committed by motorcycle riders. The system will specifically target three types of violations: the absence of helmet usage, riding in the wrong lane (going against the normal traffic flow), and engaging in triple riding. The system will use ML models (Dectectorn2 and YOLOv7) to extract motorcycle riders from video frames. Using the riders we determine triple riding and the absence of a helmet. Along with this, our tracking algorithm will create a unique identifier for each rider by linking together multiple frames of it, enabling us to identify the bike trajectory for wrong lane violations. By analyzing frames, the system can confidently issue a penalty by reading the license plate and identifying the owner of the motorcycle. Along with this, we do an extensive comparison of the Detectron2 and YOLOv7 models. However, we should consider this system’s broader implications and approach it cautiously.