一种改进的meanshift目标跟踪算法

摘要

针对传统meanshift跟踪算法不能有效消除目标内包含的背景信息、 不能自适应连续视频序列中背景的明显变化,以及不能解决光照变化带来的目标颜色特征信息变化的问题.文章提出在计算目标和背景模型直方图时,通过比较两者特征直方图的bin值,得到目标特征显著性大小,并将其代入传统的相似性度量中,同时加入一种简单的背景和目标更新算法,在有效的提高目标与背景区分度的同时,减小了因光照导致的目标直方图模型表达的偏差. 实验结果表明,该算法能在不增加计算复杂度的前提下,拥有更高的定位精度,能够有效地消除背景对目标跟踪的干扰,同时能够适应背景的缓慢变化,对光照变化也具有一定的鲁棒性.%Considering that the traditional meanshift tracking algorithm can not effectively eliminate the background information, can not self-adaptive the obvious changes of background in video sequences, and can not solve the problem of the target color characteristic information changes due to the illumination variation. This paper put forward a method that by comparing the feature histogram bin value between target and the background feature model during their computing, getting the characteristics of significant size of the target, and integrating it into the similarity of the traditional metrics, while adding a simple updating method of the background and the target. Effectively improving the difference of the target and the background area, reducing the deviation of target histogram model expression due to the changes of light. The experimental results show that, without increasing the complexity of calculating,the algorithm gets higher positioning accuracy, it can effectively eliminate the background interference of target tracking, and can adapt to slow changes in the background, certain robustness has got due to the change of light.