Helmet Rule Violation Detection for Motorcyclists using a Custom Tracking Framework and Advanced Object Detection Techniques

The use of helmets by motorcyclists is an effective way to reduce the risk of head injuries and fatalities in case of accidents. However, many countries still face the challenge of enforcing the helmet rule and ensuring compliance among riders. In this paper, we propose a novel framework that can differentiate between the driver and passengers and detect helmet rule violations for each rider empowered by computer vision and deep learning techniques. In the real-world scenario, there are many small and obstacle objects in each frame, which is a significant challenge, even with state-of-the-art detectors. To address this challenge, we employ an additional head detection module and a custom tracking algorithm that leverage auxiliary information such as moving direction, to improve detection performance on small and obstacle objects. This solution results in a significant improvement of 16% on mAP. Our complete framework achieves a final score of 69.97% on the 2023 AI City Challenge - Track 5 [18] and ranks third among the competing teams.