感兴趣区域中的快速人脸区域分割与跟踪方法

为解决人脸区域分割和跟踪的方法在速度和精度上存在的问题,提出了同时使用前景提取与肤色检测的视频监控中感兴趣区域的提取方法,以及在此基础上使用水平集方法的人脸分割和跟踪方法.在大幅度缩小AdaBoost人脸检测方法的搜索区域的同时,能够避免传统水平集方法解偏微分方程的高计算复杂度,并且在脸部朝向变化的时候保持鲁棒性.应用到PTZ摄像系统中实现视频监控系统,能够随着人脸的移动改变摄像机视角,从而在较大的区域中连续跟踪人脸.实验表明在鲁棒性和实时性上均有提高.

A Method of Fast Face Segmentation and Tracking in ROI

To enhance the speed and accuracy for segmenting and tracking face regions, an improved Regions of Interest （ROI） extraction method with foreground extraction and skin detection done at the same time in video surveillance is proposed, and level-set method used in face segmentation and tracking method. The approach is able to reduce the searching area for faces in AdaBoost face detection significantly, avoid the computational complexity of level-set solution to partial differential equations, and keep the robustness when face orientation changes. The surveillance system thus treated has a PTZ camera changing angles with face movements and can therefore track faces contiguously in a large area. Experiments show that its robustness and speed are both improved.