基于背景建模和动态分块的目标跟踪

MeanShift算法在做目标跟踪时不能解决混乱背景和严重的局部遮挡的问题。因此,本文提出了一种新的目标跟踪算法来解决这两个问题。该算法首先利用最大化后验概率在MeanShift跟踪框架内融入背景信息,然后在跟踪过程中根据候选区域和模板区域的相似度结合空间金字塔技术对目标进行动态分块,使得目标能够克服背景的干扰并保留一定的几何结构信息。实验结果表明,相比于MeanShift算法,新提出的算法能够克服混乱背景和严重的局部遮挡,有效地跟踪目标。

Background Modeling and Dynamic Partition based Object Tracking

Mean shift cannot solve the issues of confusing background and severe partial occlusion in target tracking.In this paper,a new tracking algorithm is proposed to solve these two issues.First,background information around the target is introduced into Mean Shift with a MAP formulation.Then,the target is partitioned into hierarchical blocks according to the similarity between the target candidate and target model combined with Spatial Pyramid technique.In this way,the target can overcome the interference of ba...