Abstract: In order to solve the overlapped marker matching issues and satisfy speed requirement in a 3D video motion capture system，a novel algorithm was proposed.We calculated the mean vertical coordination difference of these two parallel structure cameras to reduce 3D matching error，and the epipolar geometry restriction was used to make the corresponding matching marker points be restricted within a given area.To compare with traditional epipolar geometry restriction method that searched the matching point in a whole line，the algorithm restrict the searching region around one estimate point.Consequently，the matching speed was improved greatly owe to the reduced the searching area.To deal with the case of the points overlapped with each other，we used clustering algorithm and minimized the weighted distance to obtain the accurate matching points.Therefore，the matching accurateness has been improved.Several experimental results show this algorithm is effective.

Key words: stereo matching, epipolar line restriction, motion capture, marker points

摘要： 针对三维运动捕捉系统标记点匹配中存在标记点交叠等情况和速度要求，提出利用平行相机结构，使用统计匹配像素点的平均高度差计算相机间的高度差，并将其与外极线约束相结合将匹配标记点搜索范围限制在一个特定区域内，与传统的利用外极线约束需要在整条直线上进行搜索相比，将搜索范围限制在一个估算点周围，减小了搜索区域，提高了匹配速度。对于标记点存在交叠的情况，使用聚类的方法和加权距离最小值得到正确的匹配标记点，从而提高了立体匹配的精度。实验结果验证了算法的有效性。

关键词: 立体匹配, 外极线约束, 运动捕捉, 标记点