YOLOv5-based fatigue state detection method

Abstract

With the development of social economy and the continuous increase of car ownership in China, fatigue driving is one of the problems people need to focus on nowadays, however, most of the current fatigue state detection methods are affected by the problems of expensive, susceptible to the environment and complicated methods. In order to solve the above problems, this paper proposes a fatigue state detection method with YOLOv5m as the basic network model, which first enhances the original image and then improves the loss function. The experimental results show that the mean accuracy of this method can be as high as 95.6%, which is 4.13 percentage points higher than YOLOv4 and 6.2 percentage points higher than YOLOv3, and the model accuracy is as high as 98.27%, which is 4 and 5.2 percentage points higher than YOLOv4 and YOLOv3, respectively. The recall rate is 95.1%, which is 2 and 3.2 percentage points higher than YOLOv4 and YOLOv3, respectively. It proves the reliability and advantages of the method in this paper.