基于张量梯度直方图的Mean Shift跟踪算法  

Mean Shift Tracking Algorithm Based on Tensor Gradient Histogram

作　　者：李光[1] 金学波[1] 

机构地区：[1]浙江理工大学信息电子学院,杭州310018

出　　处：《浙江理工大学学报（自然科学版）》2009年第6期880-885,896,共7页Journal of Zhejiang Sci-Tech University(Natural Sciences)

基　　金：国家自然科学基金项目(60674028)

摘　　要：使用基于颜色直方图的Mean Shift跟踪算法,当目标在相似颜色背景、光照变化的情况下,视频跟踪性能迅速下降。文章提出了一种基于张量梯度直方图的Mean Shift跟踪算法,该算法使用张量梯度将RGB三维空间融合为一维空间,在有效地保护原向量方向性的前提下降低了维数。实验结果表明,基于张量梯度直方图的MeanShift跟踪算法耗时少,对光照变化不敏感,且对颜色变化依赖程度低,适合背景混淆和光照变化场合下跟踪。The tracking performance of the Mean Shift algorithm based on color histogram declines rapidly when the object is similar to the background color, light changes. This paper develops a Mean Shift tracking algorithm based on tensor gradient histogram, which combines the three-dimensional space of gradient tensor RGB into a one-dimensional space, in the effective protection of the original vector direction. The experimental results show that, the developed method is less time-consuming and not sensitive to light and color changes. The algorithm here is more suitable for the tracking on the condition of confusion and occasions.

关 键 词：目标跟踪 Mean SHIFT算法 张量梯度直方图 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]