Automated Traffic Monitoring and Violation Detection Using YOLOv9 and Deep Learning- Based OCR

Road safety remains a critical challenge in rapidly urbanizing Indian cities, with helmet non-compliance being a leading cause of motorcycle-related fatalities. According to the Ministry of Road Transport and Highways (2023), over 68% of two-wheeler accident fatalities involve riders without helmets. This paper presents an automated real-time traffic violation detection system using the YOLOv9 deep learning architecture with Programmable Gradient Information (PGI) and Generalized Efficient Layer Aggregation Network (GELAN). The system simultaneously detects motorcyclists, classifies helmet compliance, localizes license plates, and extracts vehicle registration numbers via advanced Optical Character Recognition (OCR). The complete violation record—including timestamp, snapshot, and vehicle registration—is automatically logged to a centralized database. The system achieves 96.9% mAP@50, processes frames at over 25 FPS, and attains 94.1% OCR accuracy under clear conditions. Results demonstrate a 4.9% improvement over YOLOv8-based baselines, with superior small-object detection enabled by the PGI mechanism. The framework offers a scalable, cost-effective solution for smart city traffic enforcement.