一种考虑驾驶员姿态及光照因素的疲劳驾驶稳健检测算法

考虑疲劳驾驶检测过程中容易出现的偏头情形和光照因素影响,提出一种疲劳驾驶稳健检测算法。在实现Adaboost和主动性状模型相结合的人眼定位基础上,算法首先通过归一化人脸图像旋转等手段,使检测系统可以适应驾驶过程中经常出现的驾驶员偏头情形;其次通过补充人脸训练样本和引入直方图均衡等手段,使其可以更好地适应驾驶中出现的各种光照环境。最后利用Perclos算法对驾驶员疲劳状态进行判定。对模拟视频及车内采集的真实驾驶样本视频进行检测实验,结果表明稳健检测算法可以更准确定位人眼的位置。不仅可以有效的适应偏头情况,并且可以消除光照因素的影响,提升了检测系统的稳健性。

A Robust Detection Algorithm for Fatigue Driving Considering Driver's Posture and Illumination

Taking two factors, head-leaning and illumination which often affect the fatigue detection of drivers, into consideration, a robust fatigue detection algorithm is proposed. On the basis of eyes location which combines Adaboost and the active shape model, the whole system is able to adapt itself to the condition of head-leaning by normalizing the spinning pictures of people＇s faces. Then the system can better adapt to various illumination conditions by supplementing training samples of people＇s faces and introducing histogram equalization. At last, the Perclos algorithm is used to judge the fatigue condition of the driver. Additionally the simulation videos and the actual driving videos are detected in the experiment and the result indicates that this robust algorithm can allocate people＇s eyes more precisely. Above all, the algorithm can not only adapt to the condition of head-leaning but also can eliminate the influence of the illumination, thereby improving the robustness of the detection system.