摘要： 序列图像中运动目标尺度在发生变化时，难以被精确跟踪。为此，提出一种基于目标直方图主成分变化检测的跟踪算法。综合考虑目标与背景的特征显著性和相似性2个因素，建立比值关系进行量化分析，选取能够最大程度区分前景目标和背景的特征，并将区分度作为权值确定直方图中所选特征的区间数。通过分析目标直方图中主成分特征分量所占的总像素数，确定目标尺度变化情况。定义一种新的带宽计算准则，自适应调整跟踪窗大小。实验结果表明，该算法能够准确地对目标尺度变化进行检测。

关键词: 图像目标跟踪, Mean Shift算法, 主成分, 自适应带宽, 特征显著性, 特征相似性

Abstract: Aiming at the problem of moving target tracking difficult under scale changes, an adaptive bandwidth algorithm is proposed by detecting the principal component variation of histogram. Considering the conspicuousness and similarity of target and background, a ratio relation is set up to select the feature which can differentiate the prospect target and background to its extent. The feature district amount is weighted by the discrimination. Target scale variation is determined by pixel amounts of principal component in histogram. A novel bandwidth criterion is defined for adjusting the tracking window adaptively. Experimental results show that the algorithm can detect the scale change of the target and make corresponding adaptation.

Key words: image target tracking, Mean Shift algorithm, principal component, adaptive bandwidth, feature conspicuousness, feature similarity

中图分类号: 

• TP391