基于肤色和Gabor纹理的粒子滤波人脸跟踪

Face Tracking Using Particle Filtering Based on Skin Color and Gabor Texture

基于单一颜色信息的跟踪方法容易受到相似颜色的干扰,应用于复杂场景时存在局限性。为此,提出一种在粒子滤波框架中结合肤色和Gabor纹理信息的人脸跟踪方法。从视频序列中提取目标人脸区域的肤色直方图以及Gabor纹理特征向量,通过这两种观测特征计算粒子集权重,估计系统状态。采用民主融合策略自适应调整观测特征的融合权重,从而增强目标描述的可靠性。同时利用彩色视频中可进行肤色检测和Gabor滤波器组在多尺度、多方向上提取纹理的优势,以及粒子滤波器能够适应非高斯、非线性系统的特点,提高视频人脸区域的跟踪精度。实验结果表明,该方法对于类肤色区域、复杂纹理背景、目标遮挡和快速移动等干扰具有较强的鲁棒性。

Tracking method based on single color information may suffer interference from similar color. There exist limitations when applied to complex scenes. Aiming at this problem, this paper proposes an algorithm combining skin color and Gabor texture in the basic frame of particle filtering. It extracts skin color histogram and Gabor texture eigenvector for face region from video frequency, then calculates particle weights and estimates syslem state. Democratic integration strategy is applied to adaptively adjust fusion weights of these observational features, thus to enhance higher target description reliability. Better tracking accuracy can be achieved by taking advantage of color video skin detection, multi-resolution, multi-orientation texture extraction of Gabor filter group, and particle filter＇s adaptability to non-Gaussian, nonlinear systems. Experimental results show this method is robust to overcome interference of irrelevant skin color region, complex texture background, target occlusion or fast moving.