中文摘要：

基于运动段分类的思想,提出一种人体轮廓提取方法.首先通过离散轮廓演变（DCE）提取场景各像素点所对应的运动段,然后在特征子空间（PCA）内对图像包含的运动段进行分类,最后去除属于背景类的运动段得到人体轮廓.在场景包含了其他运动干扰（如车辆、树叶、雨滴等）的情况下,该方法也能够高效地分割出人体轮廓区域.对USF室外步态序列的实验结果表明,经过运动段分类之后的人体轮廓提取效果较分类之前有明显改善.

英文摘要：

A novel silhouette extraction method was proposed which is completed by using motion slice classification. Motion slices were extracted from sequence of each pixel in the scene by DCE （discrete contour evolution）. Then they are classified in the feature subspace （PCA）. Human silhouette was extracted by deleting the motion slices belonging to the background. The proposed method extracts human silhouette efficiently in complicated scenes, which may include factors like vehicles, waving leaves and rains. Experimental result shows that it achieves better performance than those without motion slice classification on USF Human ID gait database.