基于多特征数据融合的疲劳驾驶检测研究

Research on fatigue driving detection based on multi-feature data fusion

为了载系统对驾驶员疲劳驾驶进行预警,从而降低交通事故发生的频率,结合驾驶员处于疲劳状态时的面部特征表现,融合多方数据来综合对疲劳驾驶进行检测.利用方向梯度直方图(HOG)特征对人脸进行检测,然后通过梯度提升决策树算法(GBDT)来获取面部的68个特征点,再利用PERCLOS算法来计算眼和嘴部的疲劳值,通过3D人脸匹配的方式来获取驾驶员头部运动姿态角度,最后通过支持向量机(SVM)算法以及训练模型对驾驶员的眼睛,嘴巴和头部姿态的特征进行融合训练来给出疲劳的综合判断.实验结果验证此检测方法能够准确的判断出驾驶员是否出现疲劳驾驶,而且时效性也得到了一定的保障.

In recent years,with the rapid development of China′s automobile field,the incidence of traffic accidents is also increasing frequently.A large part of the reason is fatigue driving.If the on-board system can give early warning of the driver′s fatigue driving,it can well reduce the frequency of traffic accidents.Therefore,this paper will combine the facial features of the driver in the fatigue state,and fuse multi-party data to detect the fatigue driving.In this method,firstly,the histogram of oriented gradient feature is used to detect the face,then the 68 feature points of the face are obtained by the gradient boosting decision tree,and the fatigue values of the eyes and mouth are calculated by the PERCLOS algorithm.Then the driver′s head pose angle is obtained by 3D face matching.Finally,the features of driver′s eyes,mouth and head posture are fused by SVM algorithm and training model to give a comprehensive judgment of fatigue.According to the experimental results,it can be verified that this test method can accurately judge whether the driver is fatigued,and the timeliness is also guaranteed.