行为分析已经成为计算机视觉研究领域的热点，行为主体的肢体部件分割是行为分析中很重要的一部分同时也是一个难点问题，为了对3维肢体进行有效分割，提出了一种基于半监督聚类的肢体分割算法。该算法首先利用前一帧姿势估计反馈的时域信息来对3维主体进行初始的肢体部件分割；然后根据人体结构信息进一步确定行为主体上各个点与肢体部件之间的关系来得到半监督聚类的初始值；之后基于各个肢体部件的形状信息进行半监督聚类，迭代求解肢体部件分割的最优解；最后利用分割后的各个肢体部件进行行为主体的姿势参数估计。通过对IXMAS 数据库中6

Human activity analysis is receiving increasing attention from computer vision researchers. One challenge is the segmentation of human body into meaningful body parts. A semi supervised clustering based body parts segmentation algorithm of 3D reconstructed human is presented in this paper. Firstly, we segment human body parts with the help of posture parameters of the previous frame. Then the structure information of human body is adopted to classify some points and initialize the centers of the semi supervised clustering. Finally, based on the shape of body parts, semi supervised clustering method is used to segment the body parts. In addition, body posture parameters are estimated with the segmentation result of body parts. The system is validated with IXMAS database, which includes 6 actors and 6 kinds of activity. The experimental results show that the presented algorithm can adapt the variety of views, actors and activitiesy.