以颜色和形状直方图为线索的粒子滤波人脸跟踪

Particle Filter Face Tracking Using Color and Shape Histogram as Clues

跟踪器的设计和跟踪线索的选择与表达是人脸跟踪中的两大关键因素,针对一般人脸跟踪算法中常用简单椭圆来描述人脸形状线索时易受背景干扰的缺点,以及视频目标跟踪中动态模型和观测模型的非线性非高斯特点,提出了一种以颜色和形状直方图为线索的粒子滤波人脸跟踪算法,该算法在粒子滤波基本框架之下,引入了一种新的用直方图来描述人脸形状的方法,并对其进行了改进,用来作为人脸跟踪的形状线索。同时,为了减轻背景干扰,提出了一种经验有效边缘的检测方法。实验表明,该跟踪方法不仅能有效地处理人脸旋转、背景中的肤色干扰和部分遮掩问题,并且能够在由于大面积遮掩等原因而丢失目标的情况下,及时有效地重新捕获已丢失的目标。

Face tracking problem comprises two key factors： one is the design of tracker; the other is the selection and representation of tracking cues. To relieve the shortcoming that the general face tracer is frequently disturbed by background when a simple ellipse model is used to describe the facial shape, and in accordance with the fact that the dynamic model and observation model of video object tracking problem are always non-Gaussian and non-linear, a robust particle filter face tracking algorithm which uses shape and color represented in the form of histogram as clues is proposed. The tracking algorithm is based on particle filter, in which a new shape representation method using histogram is introduced and modified to be as one of tracking clues. At the same time, an experiential edge detection algorithm is proposed in order to alleviate background disturbance. Experimental results show that the proposed algorithm is robust to the rotation of face, background skin color distraction and partial occlusion. In addition, the proposed tracker can recover the lost terget due to the severe occlusion.