Helmet AND Number Plate Detection System: An AI-Powered Approach to Automated Traffic Rule Enforcement

Abstract

Automated traffic enforcement is a cornerstone of smart city transportation systems, yet balancing real-time multi-object detection with high-accuracy character recognition remains a localized challenge for mixed-traffic environments. This study presents an integrated, edge-optimized deep learning framework designed for automated helmet non-compliance detection and concurrent Automatic Number Plate Recognition (ANPR). Moving away from traditional labor-intensive manual surveillance , the proposed framework utilizes a fine-tuned, 92-layer Convolutional Neural Network (CNN) based on the YOLOv8 architecture to simultaneously detect and track four distinct classes: Bike, Helmet, No-Helmet, and Number-Plate. When a multi-class violation trigger isolates a rider without a helmet, the system crops the designated vehicle coordinate matrix and routes the region of interest (ROI) to an integrated Optical Character Recognition (OCR) engine for alphanumeric text extraction. Validated on a diverse traffic surveillance dataset using a Tesla T4 GPU infrastructure, the 25.84-million-parameter model achieves an overall mean Average Precision mAP of 97.7% across all target classes. Specifically, class-isolated precision reaches 99.6% for helmet detection and 95.8% for license plate localization. Crucially, the entire execution pipeline maintains an ultra-low inference latency of 12.0 ms per frame, demonstrating a robust, high-throughput solution tailored for seamless integration with existing smart city camera networks.