Research on a DSM-YOLO-Based Detection System for Dangerous Driver Behavior in Auto-Driving

In response to the need for real-time detection of dangerous driver behaviors such as yawning, using a mobile phone, smoking, and sleeping in autonomous driving scenarios, this paper proposes a detection model named DSM-YOLO, which is based on YOLOv11 and an improved multi-scale ECA attention mechanism. Building on the efficient detection advantages of the YOLO series, have designed an improved Multi-Scale Efficient Channel Attention mechanism (MSECA) to enhance the model's capability in perceiving low-contrast, occluded, and multi-scale targets. Experimental results demonstrate that the improved DSM-YOLO significantly enhances precision, recall, and mean average precision, while also meeting the low-latency requirements for in-vehicle real-time detection.