RAY-Net: A Motorcycle Helmet Detection Method Integrated Auxiliary Correction

In the field of urban traffic safety monitoring, detecting whether the driver of a moving motorcycle is wearing a helmet is a crucial and practically significant task. However, this field currently faces two major challenges. Firstly, there is a lack of a comprehensive open-source dataset that encompasses challenging scenarios such as nighttime and rainy days. Secondly, the detection of motorcycles in motion is often disrupted by pedestrians and parked motorcycles on the roadside, making it challenging to differentiate between them and impacting the accuracy of detection outcomes. To address these issues, this paper collected and annotated images from challenging scenarios such as nighttime, resulting in 52,800 annotated images and constructing a more comprehensive dataset named the Enhanced Motorcycle Helmet Detection Dataset (EMHDD). Additionally, this paper proposes a two-stage model called RAY-Net with integrated auxiliary correction. This model includes a detection phase and a recognition phase. In the detection phase, the model tackles the problem of suboptimal detection results caused by pedestrians and parked motorcycles on the roadside through auxiliary correction. Subsequently, the recognition phase identifies the results of the detection phase and determines whether the driver is wearing a helmet.