摘要： 提出了利用人体区域彩色聚类和人体运动的相关知识，理解机器视觉中人体运动的方法。根据人类视觉区分颜色的特性和实时视觉图像，自动调整区域颜色聚类的阈值；对视觉图像中的人体进行区域聚类，并分别计算出各区域聚类中心和聚类区域的质心；根据视觉图像中前后帧相对应的聚类区域质心位置变化，利用人体运动与聚类区域质心位移关系的知识，从整体或细节上理解人体的运动。

关键词: 色彩空间；聚类中心；聚类区域质心；视觉知识；运动理解

Abstract: An approach of using human region color clustering and human motion knowledge to understand human motion in machine vision is introduced. Auto-thresholds are ascertained according to the characteristic of human classifying color and the condition of current vision image. Human motion image is then converted into HSV space from RGB space. After its color regions are clustered, the centers and the centroid of those clustered regions are computed. According to the displacement changes of those centroids between fore-and-aft frames, the human motion is understood using the human motion knowledge. The details of human motion are known in machine vision.

Key words: Color space; Clustering center; Centroid of clustered region; Vision knowledge; Motion understanding