运动捕获系统中基于多视觉的目标跟踪算法  

Multi-view Object Tracking for Motion Capture

作　　者：张欢欢[1] 郑江滨[1] 

机构地区：[1]西北工业大学计算机学院,陕西西安710129

出　　处：《微电子学与计算机》2009年第6期66-69,共4页Microelectronics & Computer

基　　金：国家"八六三"计划项目(2006AA01Z324);西北工业大学研究生创业种子基金项目(200850)

摘　　要：基于标记点的运动捕获系统中,由于标记点特征相似,易出现遮挡,使得跟踪难度大.针对上述问题提出了基于多视觉的跟踪算法.首先进行双目视觉立体跟踪:采用扩展Kalman进行预测,根据外极线约束剔出预测区域内的错误候选目标;然后进行多目视觉数据融合:利用基于可变阈值的最邻近数据融合算法对已获得的多个双目跟踪数据进行处理,解决了标记点遮挡、丢失问题;最后得到了标记点的三维运动数据.实验表明,提出的跟踪算法可以准确跟踪各个标记点.For the tracked multiple markers in a marker-based motion capture system, their images features are quite similar and they are occluded frequently, that make it become very difficult to track these markers at the same time. This paper proposes a novel multi-view object tracking algorithm, which includes binocular stereo tracking and multi-view tracking data fusing. In the procedure of binocular stereo tracking, the extended Kalman is applied and the epipolar geometry restriction is used to eliminate the incorrect matching. During of multi-view tracking data fusing, a nearest neighbor data fusion algorithm based on adaptive threshold is proposed to process these binocular tracking data, that solves the problem of occlusion and missing effectively, and then 3D tracking data are gained. Experiments show that the proposed algorithm can track multiple markers correctly.

关 键 词：运动捕获系统 双目视觉立体跟踪 多目视觉数据融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]