作　　者： ;

机构地区： 华南理工大学土木与交通学院

出　　处： 《武汉理工大学学报（交通科学与工程版）》 2018年第3期417-421,429,共6页

摘　　要： 疲劳驾驶是道路交通事故的主要诱因之一,研究基于驾驶人疲劳状态的事故预警系统,对减少交通事故、保障驾驶人的人身安全有重大意义.文中提出一个基于卷积神经网络的驾驶人疲劳检测系统,由预处理模块和疲劳检测模块组成.预处理模块应用方向梯度直方图(HOG)从待处理图像中定位人脸区域,再运用特征点检测算法和旋转变换校正人脸姿态.疲劳检测模块使用卷积神经网络对驾驶人的面部特征进行提取,再经由网络最后一层进行疲劳判别.与传统的基于瞳孔开度等单一特征的检测方法相比,卷积神经网络会在训练的过程中自行选择特征,不依赖于人工参与,能有效对抗户外恶劣环境的干扰,克服传统疲劳检测方法在光照剧烈变化的实车环境下遇到的困难.本文对算法进行了室内测试和实车测试,在室内测试中系统取得了综合96%的准确率,在实车测试取得了87%的准确率,验证了该方法在干扰强烈的室外环境下的准确率和鲁棒性. Fatigue driving is one of the main causes of road traffic accidents.The study on accident early warning system based on driver＇s fatigue state has a great significance to reduce traffic accidents and ensure the personal safety of drivers.In this paper,based on convolution neural network a driver fatigue detection system consisting of a preprocessing module and a fatigue detection module was proposed.The preprocessing module applied the direction gradient histogram（HOG）to locate the face region from the image to be processed,and the facial gestures were corrected by using feature point detection algorithm and rotation transformation.The fatigue detection module used convolution neural network to extract the facial features of drivers,and then carried out fatigue judgment through the last layer of network.Compared with the traditional detection method based on single features such as pupil opening,convolution neural network would choose its own characteristics in the process of training,which is independent of human participation and can effectively resist the interference of outdoor harsh environment.Meanwhile,the proposed method overcame the difficulties encountered by traditional fatigue detection methods in real vehicle environments with drastic changes in light.The indoor tests and real vehicle tests of the algorithm were conducted in this paper,it has achieved the comprehensive 96% accuracy in the indoor test,and 87% accuracy in the real vehicle test.The accuracy and robustness of this method in outdoor environment with strong interference are verified.

关 键 词： 疲劳驾驶 深度学习 卷积神经网络

领　　域： []