Abstract: The lower limber joint of human body location is currently one of the most important research topics in the domain of human motion tracking and human motion analysis.A novel approach is proposed for locating lower limber joints of human body from image sequences without markers.First the method employs a background subtraction to separate human object，and builds skeleton model of human lower limbers.Second the approach uses the prediction of joint-angles to evaluate the position of knee and ankle，and captures the exact position of lower limber joints through matched computing based on appearance model.Experimental results show that the approach is simple and effective，and the locating results are satisfactory.

Key words: human motion tracking, human motion analysis, lower limbers, appearance model, joint position location

摘要： 近年来，人体下肢关节点定位成为了人体运动跟踪与分析中一个重要研究课题。提出了一种下肢关节点自动定位方法，从无关节标记的人体运动图像序列中定位下肢关节点。该方法首先采用背景剪除技术从图像序列中分割人体目标对象，并建立人体下肢骨架模型。然后，利用关节角度预测方法估计膝踝关节点的位置，在基于下肢外观模型的匹配计算基础上获得下肢关节点的真实位置。实验结果表明，该方法简单有效，下肢关节点定位结果令人满意。

关键词: 人体运动跟踪, 人体运动分析, 下肢, 外观模型, 关节点定位