Robust Motorcycle Helmet Detection in Real-World Scenarios: Using Co-DETR and Minority Class Enhancement

Motorcycle helmet detection is a crucial task in intelligent traffic systems (ITS), as it enhances traffic safety consciousness and guides individuals towards legal compliance. Numerous challenges are tied to this problem, particularly regarding data from the real world. In addition to requiring resilience to environmental fluctuations, such as diverse camera angles and lighting conditions, the solution must also address the problem of unbalanced data distribution across object classes. This study presents a system that utilizes Co-DETR to address the difficulties of dealing with changing perspectives on real-world data. Additionally, we propose to use the Minority Optimizer and the Virtual Expander to enhance the accuracy of rare classes in imbalanced data. With a mean average precision (mAP) of 0.4860, our method achieved Rank 1 in the AI City Challenge 2024 Track 5 competition, demonstrating its high effectiveness.