基于视频图像和RBF神经网络的疲劳驾驶监测系统

Driver Fatigue Monitoring System Based on RBF NN and Image Processing

采用视频采集方式和神经网络方法实现了驾驶员疲劳驾驶的非接触式监测。应用车头前端和车厢内部双路视频摄像头分别采集车辆相对于车道线的行驶轨迹和驾驶员的睁闭眼状态,应用Radon变换提取5 s内车头与车道线间的最大和最小偏离、相邻2帧间车头与车道线的最大角度变化量和平均角度差,应用AdaBoost算法提取驾驶员眼睛闭合帧数比例,并将上述各参数作为RBF神经网络的输入来实现驾驶员疲劳状态的动态监测,实验数据表明监测效果良好。

As effective un-touched driver fatigue recognize tokens,driver s per eye close（PERCLOS）and vehicle s track has been used in this paper.Two cameras fixed at vehicle head and vehicle inside are used to capture the video stream of driver s face and vehicle track.Image processing technology is used to extract 5 parameters per 5 seconds,e.g.Radon transform applied to maximum and minimum distance from vehicle head to lane line and maximum and average angle difference from vehicle central line to lane line between adjacent images from camera on vehicle head, AdaBoost algorithm applied to PERCLOS according to driver＇ s face images from camera inside. These 5 parameters act as input data for RBF NN to recognize whether driver is fatigued or not. Trail data show the method is effective.