Automated Multi-Stage Helmet Detection using Surveillance Cameras for Indian Traffic - A Deep Learning Approach

The increasing affordability of two-wheelers in India has led to their widespread adoption as a primary mode of transportation. However, many riders exhibit reluctance in wearing helmets, contributing to rising road accidents and fatalities. While existing helmet detection systems offer potential solutions, they often require human intervention, raising concerns about efficiency, reliability, and bias. Automating helmet detection can enhance compliance with safety regulations, but real-time implementation remains challenging due to diverse environmental conditions. This paper proposes a multi-stage automated helmet detection framework designed for real-time deployment in surveillance systems. The system leverages YOLO(You Only Look Once)-based deep learning models to detect vehicles, riders, and helmet usage sequentially. Extensive evaluations demonstrate that YOLOv8n and YOLO11n achieve the best trade-off between accuracy and computational efficiency (less than 0.2 seconds for inference per image), with YOLO11n offering a lower parameter count, making it more suitable for real-time applications. The detected two-wheelers are subsequently processed through a rider detection module, followed by helmet classification. Helmet violations are automatically extracted and stored in a cloud-based repository for further enforcement. The proposed system effectively minimizes the trade-off between accuracy, inference speed, and cost-efficiency, making it a practical solution for real-time helmet detection in large-scale traffic monitoring applications.