Motorcycle Helmet Detection using Deep Learning

Abstract

Motorcycle accidents account for over 23% of global road fatalities, with head injuries as the leading cause. Research indicates helmets can reduce fatal injury risk by 42–69%, yet compliance remains inconsistent due to reliance on manual monitoring. This work proposes an automated computer vision solution using YOLOv8, trained on 400 annotated images (classes: helmeted riders, non-helmeted riders, occupants). The model achieves 0.677 mAP and 0.67 F1-score, demonstrating viability for CCTV-based traffic monitoring. While further optimization could improve robustness, this approach enables 24/7 enforcement without infrastructure costs