具有自适应MB-LBP前置滤波的CAMShift人脸跟踪算法

基于颜色分布的连续自适应均值移动（CAMShift）人脸跟踪算法简单、易于实现，被广泛应用于实时跟踪。但因其采用肤色模型作为跟踪模式，所以当目标处于类肤色背景区域时，跟踪窗口极易错误收敛到背景区域从而导致跟踪失败。为此，本文提出一种具有自适应LBP前置滤波的CAMShift跟踪算法。首先训练一个能检测人脸基本特征的级联MB-LBP节点分类器。当跟踪窗口进入类肤色干扰区时，系统自适应地把该分类器接入作为CAMShift跟踪算法的前置滤波器，以排除背景中的类肤色干扰，提高算法的鲁棒性。实验结果表明，该算法既能有效排除背景中的类肤色干扰、显著提高CAMShift人脸跟踪算法的鲁棒性，又能保持人脸跟踪的实时性。 

CAMShift face tracking with adaptive MB-LBP pre-filter

Since the color probability distribution based continuous adaptive mean shift (CAMShift) face tracking algorithm is simple and easy to be implemented, it is widely used in real-time tracking applications. However, because CAMShift takes skin color histogram as the tracking model, its tracking is easy to fail when the target locates in a skin-color-like background region. For that, a CAMShift face tracking algorithm using adaptive Multi-Block Local Binary Pattern (MB-LBP) pre-filter is proposed in this paper. Firstly, a cascade MB-LBP node classifiers, which can well detect the basic characteristics of face, is trained. If the tracking window enters a skin-color like background region, this pre-filter is then adaptively inserted to eliminate the skin-like-color background interferences. Consequently, the robustness of the tracking algorithm is improved. Experimental results have proved the superior tracking ability of the proposed algorithm under skin-color-like background interferences.