Night Fatigue Driving Detection Algorithm based on Lightweight Zero-DCE

Abstract

Aiming at the problem of low image exposure in low-light scenes at night, resulting in low accuracy of fatigue driving detection, a lightweight Zero-DCE night fatigue driving detection algorithm was proposed The depthwise separable convolution is used in the backbone feature extraction nebv0rk of the Zero-DCE model to improve the speed of the detection nebv0rk and reduce the amount of nebv0rk parameters; the down-sampled input is used as the input of the enhanced nebv0rk, and the output is mapped back to the original resolution by up-sampling. Perf0rm image enhancement, effectively balancing enhancement performance and significantly reducing computational cost. The facial eye and mouth features are detected by the target detection algorithm and the open and closed states are identified and the detection results are calculated and output according to the eye and mouth fatigue parameters combined with the threshold The experimental results show that in the low-light environment at night, the detection algorithm proposed in this paper improves the detection accuracy by 17.07% compared with the existing algorithm, and the detection time after algorithm fusion is 0.012s, which is more in line with the application requirements of fatigue driving detection scenarios.