Lightweight fatigue driving detection method based on improved Yolov8

Fatigue driving is one of the main factors leading to traffic accidents. Accurate detection and early warning of this behavior is of great significance to road traffic safety. Most of the existing deep learning fatigue driving detection methods focus on improving accuracy, ignoring detection speed and algorithm volume, and can not be used in vehicle equipment with limited computing resources. Therefore, this paper designs a lightweight network model based on the target detection network Yolov8 to speed up fatigue detection. The C2F module in the backbone feature extraction network CSPDarknet53 of Yolov8 is replaced by the block module in MobileNetv3. The feature extraction speed is accelerated and the number of parameters and calculation amount are reduced. Then, the SE attention mechanism is introduced into the Neck structure. In the output part, the loss function Wise-IoU with dynamic non-monotonic focusing mechanism is used to improve the accuracy of detection. Finally, through comparative analysis, it is concluded that the speed of the improved Yolov8 model reaches 109.81 frames per second, and the average accuracy reaches more than 95%. Based on meeting the practical application, its universality is greatly improved, and it can better adapt to the fatigue driving detection environment.