Enhancing YOLOv7-Based Fatigue Driving Detection through the Integration of Coordinate Attention Mechanism

Driver fatigue is a significant cause of traffic accidents, necessitating the development of effective methods for detecting and preventing drowsy driving. This paper presents a novel and enhanced fatigue driving recognition network, referred to as CA-YOLOv7, which inserts the Coordinate Attention mechanism between neck and head in the original YOLOv7 network structure. Additionally, it is combined with the PERCLOS algorithm to determine the level of fatigue. Comparative experiments were conducted on a self-collected fatigue driving detection dataset in VOC format to assess the effectiveness of the CA-YOLOv7 model in comparison to YOLOv7, SE-YOLOv7, and ECA-YOLOv7 models. The results demonstrate that CA-YOLOv7 has the best performance in experiments, increasing the YOLOv7 mAP value from 99.10% to 99.30% and the detection speed from 53.33 FPS to 69.44 FPS. This experimental results indicate that CA-YOLOv7 exhibits outstanding performance concerning both accuracy and speed for recognition. Overall, the proposed method can effectively guarantee driving security by detecting and preventing driver fatigue.