An Effective Motorcycle Helmet Object Detection Framework for Intelligent Traffic Safety

Shun Cui, Tiantian Zhang, Hao Sun, Xuyang Zhou, Wenqing Yu, Aigong Zhen, Qihang Wu, Zhongjiang He; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023, pp. 5470-5476

Abstract


Detecting violations of motorcycle helmet rules is an important computer vision task that can greatly protect the lives of motorcycle drivers and passengers in traffic accidents. This abnormal event detection problem can be viewed as an image object detection task, which aims to detect the location of the motorcycle driver and passenger in the image and whether they are wearing helmets. In this paper, we propose a motorcycle helmet object detection (MHOD) framework to achieve this task. Specifically, we first utilize the object detection network with ensemble model to predict the location and category of all objects in videos which can improve the accuracy and robustness of detection model. Then for the scarcity of passenger category training data, the Passenger Recall Module (PRM) is designed via tracking refinement which greatly improves passenger category recall. Finally, we introduce the category refine module (CRM) to correct the category by combining the temporal information in the video. On the test dataset of AI City Challenge 2023 Track5, we achieve significant result compared with other teams, the proposed model ranks first on the public leaderboard of the challenge.

Related Material


[pdf]
[bibtex]
@InProceedings{Cui_2023_CVPR, author = {Cui, Shun and Zhang, Tiantian and Sun, Hao and Zhou, Xuyang and Yu, Wenqing and Zhen, Aigong and Wu, Qihang and He, Zhongjiang}, title = {An Effective Motorcycle Helmet Object Detection Framework for Intelligent Traffic Safety}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, month = {June}, year = {2023}, pages = {5470-5476} }