基于肤色的驾驶员面部定位与跟踪算法

Complexion-Based Algorithm for Driver's Face Locating and Tracking

采用CCD摄像机获得驾驶员脸部图像,利用肤色在颜色空间的聚类性,建立基于肤色的人脸二维高斯分布数学模型.选用YCbCr颜色空间作为肤色分布统计的映射空间,用最大类间方差阈值对人脸图像分割,获得人脸图像灰度直方图水平投影和垂直投影,实现对面部定位.采用卡尔曼滤波算法,根据驾驶员面部当前帧图像预测下一帧图像的位置,并通过局部搜索精确定位,实现对驾驶员面部的跟踪.实验结果表明,基于肤色的驾驶员面部定位与跟踪算法可靠性高,具有良好的动态定位能力,而且对驾驶员坐姿的变化有较好的适应能力.

A 2-dimensional mathematical model of Gause distribution was developed on the basis of human face complexion to obtain driver＇s face images using CCD camera and by the clustering analysis in chrominance space. It selects chrominance space as the statistical distribution of complexion and segments face image by maximum intercluster variance threshold to get horizontal and vertical projection of histograms of gradation of face image, thus locating the face. Kalman filter is used to track driver＇ s face by predicting the location of image in next frame in accordance to the face image of driver in present frame and locating it precisely through local search. The experiment results showed that the algorithm to locate and track driver＇s face on the basis of complexion has high reliability, dynamically locating ability and adaptability to driver＇s posture.