Edge–Cloud Integrated Real-Time Helmet Violation Detection Framework Using YOLOv5 for Intelligent Traffic Monitoring

Abstract: Road safety enforcement remains a persistent challenge in regions with high motorcycle density, where manual monitoring of helmet compliance is inefficient and error-prone. Although recent deep learning models have improved helmet detection accuracy, many existing systems are limited to standalone detection without scalable deployment or automated violation management. This paper presents an edge–cloud integrated real-time helmet violation detection framework based on the YOLOv5 object detection architecture. The proposed system processes traffic video streams at the edge for rapid inference while synchronizing violation events to a cloud layer for centralized monitoring and alert generation. A custom annotated dataset was used to fine-tune pre-trained YOLOv5 weights with optimized hyperparameters and data augmentation strategies. Experimental evaluation demonstrates a mean average precision (mAP@0.5) of 94%, precision of 0.92, and recall of 0.90, with inference speeds exceeding 30 FPS, confirming suitability for real-time traffic scenarios. The integration of cloud-based violation logging enhances scalability and supports intelligent traffic analytics. The proposed framework provides a deployable and extensible solution for automated helmet compliance monitoring in smart transportation systems.