A fatigue driving detection approach based on TPU computing device

Fatigue driving is a longstanding hazard in the field of road traffic safety. The factors such as longtime driving, lack of sleep, and work pressure result in driver fatigue which impair driver ability to detect and respond to sudden situations, and increase the risk of serious traffic accidents. This paper proposes a novel method for driver fatigue detection based on YOLOv5 algorithm. Firstly, a fatigue detection dataset is collected and created for this purpose. The YOLOv5 algorithm is then utilized to detect driver multiple facial fatigue features such as yawning, closed eyes, and head nodding, and the driver's fatigue status is determined based on the statistical frequency of the three fatigue characteristics. Furthermore, the YOLOv5 trained model is deployed on the Tensor Processing Unit (TPU) computing device. The experimental results demonstrated that the YOLOv5 algorithm model achieved a mAP@0.5 of 98.51% when evaluated on the validation sets. The driver fatigue detection system deployed on the TPU computing device can perform at 25 fps with an average accuracy of 93.8%. This system can monitor the driver's status in real time in practical applications, timely remind drivers to pay attention to safety, and help reduce the risk of traffic accidents caused by fatigue driving.