一种鲁棒的均值漂移与彩色特征点提取相结合的面部跟踪算法

针对复杂背景中的多人脸跟踪问题,提出一种均值漂移与彩色特征点提取相结合的面部跟踪算法.手动选取要跟踪的面部区域,并利用彩色Harris探测器进行特征点提取,然后根据肤色的色度信息,利用均值漂移算法粗略定位人脸的位置,之后对这个区域提取特征点,与其前提取的特征点进行匹配,满足约束条件的区域即为要跟踪的人脸区域.实验表明,该算法可以排除近肤色物体的干扰并可以在多人脸发生重叠的情况下保持稳定跟踪.

Robust Face Tracking Method Based on Mean Shift and Color Interest Point Detecting

A method using mean shift and color interest point detecting is proposed in order to track faces in the complex background.We manually select the face domain,and detect interest points by the color Harris detector.We roughly fix on the face domain by mean shift algorithm based on the chrominance of the face color,and then detect color interest points in those domains.We match the color points in the two frames.Only if the points satisfy the constraint condition,the domain is determined as the face domain.Experiments results prove our method can effectively eliminate the disturbance of skin color objects and can track the faces individually which are occluded by each other.