一种点相关统计特性的目标跟踪方法.docVIP

 n 一种点相关统计特性的目标跟踪方法 武文斌 1，毋立芳1，王晓芳 2，王向东 3 （1．北京工业大学电子信息与控制工程学院，北京，100124；2. Villanova University Dept. of Electrical and Computer Engineering, Villanova, PA 19085；3.国家体育总局体育科学研究所, 北京, 100061） 摘要：提出了一种新颖、鲁棒的运动目标跟踪方法。在首帧图像建立特征点和物体中心点的 关系，然后在后续帧图像使用 Lucas Kanada 跟踪方法来跟踪这些特征点。利用这些特征点 来计算一个物体中心位置的点集，并提出一种统计最优方法来鉴别那些存在跟踪误差较大的 特征点，之后在计算中心点过程中把这些特征点剔除掉，并使用子图像匹配方法来矫正存在 跟踪误差的特征点。测试了 8 个实验视频，结果证明该算法不仅消弱了跟踪误差对中心位置 计算的影响，而且基本消除了跟踪误差积累的弊端。 关键词：多媒体技术；物体跟踪；点相关性；累积误差；统计最优；中心中点连线 中图分类号：TP37 文献标识码：A Robust object tracking approach using point correspondences Wu Wenbin1, Wu Lifang1, Wang Xiaofang2, Wang Xiangdong3 (1.School of Electronic Information and Control Engineering, Beijing University of Technology, Beijing, 100124, China; 2. Department of Electrical and Computer Engineering, Villanova University, Villanova PA 19085, USA; 3. China Institute of Sport Science, Beijing 100061, China) Abstract：We present a novel robust tracking scheme for moving objects. Firstly we build the relationship between feature points and the centre of an object in the first frame. The Lucas Kanada tracker is then used to track these feature points in subsequent frames. We use these feature points to calculate a set of values for the centre location of the object. Then we propose a statistically optimal method to identify the values with large tracking errors, excluding which the location of centre is computed. Furthermore, the location of the feature points with tracking error is corrected by using a sub-image matching method. Experimental results using eight real video data sets demonstrate that our proposed method successfully removes the impact of tracking error of a few points on the centre location. Moreover, our solution prevents the accumulation of tracking errors. Key words: image motion analysis; object tracking; point correspondences; accumulated error; statistically optimal; link_centre_midpoint (lcm) 0 引言 近年在视频监控、交通控制、体育训练、计算机控制领域，基于视觉的物体跟踪被广泛 应用，已经成为一个热门的研究方向。在某些应用场景中，有一些特征点很容易跟踪，但是