基于子空间约束的面部特征点跟踪算法

Facial feature tracking based on subspace constraint

为克服纹理不丰富和非刚性形变等因素引起的面部特征点跟踪困难,提出了一种基于子空间约束的面部特征点跟踪算法.针对人脸运动特点,将面部特征点分为具有复杂运动模式和简单运动模式的特征点集.用通过样例学习得到的特定描述模型准确刻画了具有复杂运动模式的特征点集的变化,保证了子空间约束的有效性.对运动模式简单的特征点集的跟踪则采用基于光流的算法,以提高算法的效率,也为基于特定描述模型的跟踪算法提供了更准确的起始搜索位置.对跟踪结果进一步应用子空间约束解决跟踪中的开孔问题和消除跟踪误差.实验结果表明,在存在较大脸部变形和部分特征点纹理不丰富的情况下,该方法可以有效地跟踪较密集的面部特征点.

The paper proposes a method for facial feature tracking using subspace constraint. According to the characteristics of facial motion, the facial feature points are divided into two subsets. For those features which have non-rigid deformations, we propose to track them based on the specific model learned from examples. For those features which have rigid or nearly rigid motions, we employ optical flow based method. The model-based method ensures that the constraint will not be degenerate and the optical flow based method provides a better guidance for model-based tracking. The results from the two methods are combined and refined using the subspace constraint. Experimental results show that the proposed method can efficiently track dense facial features even when some facial features have degenerated texture.