An embedded device-oriented fatigue driving detection method based on a YOLOv5s

摘要

Currently, most fatigue driving detection methods rely on complex neural networks whose feasibility in hardware implementation needs to be further improved. This paper proposes an embedded device-oriented fatigue driving detection method based on a lightweight YOLOv5s. Firstly, a YOLOv5s face detection network with a parametric-free attention mechanism is designed to enhance the focus on face regions during face detection. Then, a practical facial landmark detector model is improved by integrating multi-scale feature fusion with Ghost module, which can adapt to the variations brought by different scale targets. Next, a fatigue determination approach is investigated by using multiple features of the face. Finally, experiments of the proposed detection model with the public YawDD dataset are implemented on the PC platform and the embedded device, respectively. The experimental results demonstrate that the proposed method achieves a detection accuracy of 95.3% and a processing speed of 22FPS on the PC platform. Meanwhile, the hardware test on an Orange Pi5 embedded device achieves a detection accuracy of 93.3% and a processing speed of 12FPS, which has good prospects for applications.

关键词