基于深度学习和面部多特征融合的驾驶员疲劳检测研究

Research on Driver’s Fatigue Detection Based on Deep Learning and Facial Multi-feature Fusion

摘要:

驾驶员疲劳驾驶是造成交通事故的重要原因之一。本文提出了基于深度学习和面部多特征融合的驾驶员疲劳检测研究方法。首先采用多任务级联卷积神经网络MTCNN（Multi-task convolutional neural network)结构进行面部检测和特征点定位，并利用Dlib工具包中的面部68个地标，提取驾驶员面部的特征参数；其次，基于眼睛纵横比（EAR），眼睛闭合百分比（PECLORS）和嘴巴高宽比（ MAR）的值按不同的权值相加得到参数M,在一定时间内累加M>0.605的帧数判断驾驶员疲劳的程度。最后，试验结果表明：本方法能够有效地利用视频图像实时检测驾驶员疲劳状态，其准确率和灵敏度分别为93.1%和90.2%，对于保护驾驶员及车辆行驶安全具有重大意义。

Abstract:

Driver fatigue driving is one of the major reasons of causing traffic accidents. This article puts forward a detection method of driver’s fatigue based on deep learning and facial multi-feature fusion. First, the Multi-task convolutional neural network to conduct face detection and feature point location, and makes use of 68 landmarks on the face of Dlib toolkit to extract the characteristic parameters of the driver's face; Secondly, it adds the different weights to get the parameter M based on the value of eye aspect ratio(EAR), percentage of eyelid closure over the pupil over time(PERCLORS) and mouth aspect ratio(MAR), and accumulates the frame number of M >0.605 in certain time to judge the degree of driver’s fatigue. Finally, the experimental results indicate that: this method can make effective use of video images to detect driver’s fatigue state in time, its accuracy and sensitivity are 93.1% and 90.2% respectively, which has great significance for driver protection and vehicle safety.