Abstract: Current template-matching based face tracking algorithm can not deal with the variation of faces while tracking. This paper firstly introduces Lie groups and the model of skin tone, gets the best ellipse to approximate the face contour using affine transformations on Lie groups. Next it presents an algorithm to get rid of the dependence of initial position in the problem of face tracking on the basis of model of skin tone. At last, it proposes an algorithm to track faces in videos. The experimental results demonstrate that the algorithms proposed can detect faces very well, adapt to the variation of faces, and improve the adaptability in tracking process.

Key words: Lie group, skin tone, face detection, face tracking

摘要： 目前基于模板匹配的人脸跟踪算法并不能很好地应对人脸的多种变化。介绍了李群与肤色模型，在人脸的椭圆模型上结合李群仿射变换，获得了检测人脸的最佳椭圆，以应对人脸的不同姿态；并在该检测算法的基础上，结合肤色模型，解决了跟踪普遍存在的初值问题，给出了视频中人脸跟踪算法。实验结果表明，该算法可以很好地检测人脸，应对人脸的多种变化，并在跟踪过程中可以不断提高算法的适应能力。

关键词: 李群, 肤色, 人脸检测, 人脸跟踪