Enhancing Road Safety and Accountability Through Automated Rider and Helmet Detection: A Deep Learning Approach

This research explores the application of deep learning to address the problem of road accident severity, especially in two-wheeler riders, which occurs due to helmet non-compliance. Despite the strict enforcement of traffic laws, ensuring rider accountability needs attention to overcome the situation. Computer vision can provide a real-time solution to monitor the condition of two-wheeler riders where a system can detect riders first following the detection of helmets in their heads. In this work, a pipeline workflow of two pre-trained deep learning models is used: i) YOLOv8 for rider segmentation, and ii) VGG16 for riders with and without helmet classification. This research analyzed the successful implementation of the process with mean Average Precision (mAP) 95% from segmentation and 99.5% accuracy from classification by combining YOLOv8 and VGG16, respectively. 3301 images are used after the manually done preprocessing. The output of this model is trained to detect two-wheeler riders from the input, followed by the classification of riders with or without a helmet. The model is further trained to differentiate between a rider and a passer-by in crowded situations. The proposed system is expected to contribute to traffic management through automation of law violator detection using real-time monitoring.