On Fatigue Driving Detection System Based on Deep Learning

Aiming at the present methods of detecting human fatigue, this paper proposes a new idea of fatigue detection based on deep learning. First of all, YOLOV3-Tiny algorithm is used to detect faces and open mouths in images. Compared with SSD, FCNN and other object detection algorithms, YOLOV3-Tiny has a higher detection and recognition rate for small object, and can also detect targets faster. Then a variant based on LeNet-5 network was used to detection the closed state of the eyes. Compared with the traditional hand-crafted human eye feature descriptor, the deep learning method adopted in this paper can identify the closure of eyes more accurately and has better robustness. Finally, the improved PERCLOS algorithm is used to judge fatigue.