基于深度学习模型的疲劳驾驶行为识别算法

Algorithm of fatigue driving behavior recognition based on deep learning model

为降低道路交通事故发生率,提出了一种基于深度学习模型的疲劳驾驶行为识别算法。采用照度增强和反射分量均衡化的方法,以提高视频图像质量。将机器视觉工具箱软件用于提取疲劳驾驶人脸行为特征,并通过双流网络构建和训练深度学习模型,实现对疲劳驾驶行为识别。选择了不同睡眠时间段参与者在全封闭路段内的驾驶行为图像,作为实验测试目标。结果表明:用该算法测试1000张疲劳驾驶行为图像时,识别时间为89 ms,精准度为97.6%,召回率为97.0%;算力需求(每秒所执行的浮点运算次数,FLOPS)≤88;该算法能够提高疲劳驾驶行为的识别精度,有助于降低道路交通事故的发生率。

A fatigue driving behavior recognition algorithm was proposed based on the deep learning model to identify fatigue driving behavior and to reduce the incidences of road traffic accidents.An illuminance enhancement method and a reflection component equalization method were used to improve the quality of video images.The Machine Vision Toolbox software was used to extract facial behavior features of fatigued drivers.A deep learning model was constructed and trained by using a dual-stream network to achieve fatigue driving behavior recognition.The images of participants'driving behavior in fully enclosed segments during different sleep periods were selected as experimental test targets.The results show that the recognition time is 89 ms,the accuracy is 97.6%,and the recall rate is 97.0%when 1000 images of fatigue driving behavior are tested by the proposed algorithm;The computing power requirement(floating-point operations per second,FLOPS)is less than or equal to 88.Therefore,this algorithm improves the recognition accuracy of fatigue driving behavior,helps to reduce the incidence of road traffic accidents.