Automated Real-Time Traffic Violation Detection: An Advanced Framework Utilizing YOLOv9 and Deep OCR for Smart City Governance

Ensuring road safety in urban environments requires effective monitoring of traffic violations such as riding without helmets and vehicle identification for automated enforcement. This paper presents a robust, automated vision- based system utilizing the latest YOLOv9 (You Only Look Once v9) architecture. By integrating Programmable Gradient Information (PGI) and the Generalized Efficient Layer Aggregation Network (GELAN), the system addresses the “information bottleneck” common in deep networks, achieving superior detection accuracy for small objects like helmets and license plates in complex scenes. Extracted license plate regions are processed using high-precision EasyOCR for automated record-keeping. Results demonstrate a significant advancement over previous YOLOv8-based frameworks, offering a mean Average Precision (mAP) of 96.9% and real-time inference speeds suitable for edge-device deployment.