AI-based Helmet Detection for Motorcycle Drivers: Isfahan Traffic Control Center Case Study

Helmet detection is crucial for enhancing motorcycle rider safety and enforcing traffic regulations. In this study, we fine-tune YOLO۹, a state-of-the-art object detection model, on a custom dataset collected from the Isfahan Traffic Control Center to improve helmet detection accuracy. The proposed model demonstrates enhanced performance compared to the original YOLO۹, achieving higher precision (۷۵.۱% vs. ۷۳.۵%), recall (۷۰% vs. ۶۷%), mAP@۰.۵ (۷۲.۶% vs. ۶۹.۵%), and mAP@۰.۵:۰.۹۵ (۶۰.۹% vs. ۵۷%). A qualitative analysis further supports these findings, highlighting the model's robustness across different lighting conditions, viewing angles, and helmet types while identifying challenges related to occlusion and false detections. The results confirm the effectiveness of fine-tuning YOLO۹ for real-world helmet detection applications.

کلیدواژه ها:

Helmet detection ، transfer learning ، Object detection ، Traffic control