面向驾驶人疲劳检测的人脸视频分析算法研究

Research on face video analysis algorithm for driver's drowsiness detection

疲劳驾驶是造成交通事故的重要原因。基于面部视频分析技术,对驾驶人的眼睛动作和状态进行特征分析,可以有效估计驾驶人的疲劳状态,但驾驶过程中驾驶人面部姿态和光照条件的变化使眼睛的准确定位变得困难。本文以主动形状模型(ASM,Active Shape Model)为基础对面部区域进行配准,结合Lucas-Kanade光流算法进行全局跟踪,并采用基于自商图的Meanshift算法进行局部校准。实验结果表明,Meanshift算法能够有效消除光流全局跟踪中的累积误差,有效提高人眼定位的精度。

Driving drowsiness was considered to be major cause to traffic accidents. The driver＇s eye state monitoring based on face video analysis was an effective way to detect driving drowsiness. However. eyelocation became difficult because driver＇s pose and illunmination vary. Based on face alignment by using Active Shape Model （ASM）, this paper presented an eye tracking algorithm, combined with Lucas-Kanadeoptical flow and Meanshift. While it took full advantage of the high converging speed of Lucas-Kanade optical flow, Meanshift could eliminate the accumulated tracking errors of optical flow. The proposed algorithm exhibited good tracking performance in the experiments.