AI-Driven Surveillance for Helmet Detection and Number Plate Recognition

As the cases of two-wheelers on roads brought with it surge in helmet violations, which makes monitoring systems important that are automated. This study deals with automatic detection system for helmet violation detection with Automated Number Plate Recognition. A YOLOv11m object detection model act as backbone, dividing motorcycle riders as helmeted or non-helmeted then locating number plates within live CCTV footage. To reduce computational overhead, selective ANPR mechanism is mentioned, where Optical Character Recognition (OCR) is executed when rider without helmet is trapped. This mechanism minimizes unnecessary computation and enhances overall system efficiency. The model is trained on custom dataset of 8,460 labelled traffic images captured under different illumination, different camera angles, occlusions, and different traffic densities. Experimental results show a mean Average Precision (mAP@0.5) of 95.3%, helmet detection precision of 92.9%, and number plate detection precision of 95.5%, with real-time processing speed of 56 FPS. Comparative evaluation against YOLOv5 and YOLOv8 demonstrates the superior performance of YOLOv11m in crowded urban traffic scenarios. The proposed system can be deployed in smart city surveillance, e-challan generation, and automated law enforcement systems.