An Effective Method for Detecting Violation of Helmet Rule for Motorcyclists

Yunliang Chen, Wei Zhou, Zicen Zhou, Bing Ma, Chen Wang, Yingda Shang, An Guo, Tianshu Chu; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2024, pp. 7085-7090

Motorcycles are one of the most popular modes of transportation. However motorcycle riders are exposed to a greater risk of crashes due to fewer safety protection measures compared to drivers of cars or other standard vehicles. Thus helmet plays an essential role in protecting the safety of motorcycle riders and passengers. Nevertheless people pay little attention to it particularly in developing countries such as India. Therefore video surveillance-based automatic detection of motorcyclists without wearing helmets is one of the critical tasks to enforce strict regulatory traffic safety measures. To this end we introduce a simple yet effective method for detecting violation of helmet rule for motorcyclists in this paper. Two transformer-based detectors DETA and Co-DETR are employed to detect motorbikes and riders not wearing helmets. We enhance the generalization ability of our models using several data augmentation techniques. Furthermore we explore different fusion strategies merging the predictions from different detection models to improve the performance of our method. Our method achieves a mAP of 0.4824 on the public leaderboard of 2024 AI City Challenge Track 5 without any tracking or post-processing steps.