In road traffic safety management, not wearing a helmet has become one of the main causes of injuries and fatalities among drivers. Timely detection of riders not wearing helmets is extremely important for reducing road traffic accidents and ensuring the safety of people’s lives and property. This paper proposes the C-YOLOv5 model, which optimizes the YOLOv5 model using the CBAM attention mechanism to identify the violation of not wearing a helmet, thereby improving the accuracy and efficiency of helmet detection for non-motorized vehicle riders, providing strong support for road traffic safety management. Experimental results show that the C-YOLOv5 model has improved detection accuracy for helmet usage and localization performance for small targets, offering an effective solution for helmet detection among non-motorized vehicle riders, while also providing reference value for the research and application of object detection algorithms.