摘要
针对在相似颜色干扰场景下传统均值漂移目标跟踪鲁棒性差的缺陷,提出Prewitt梯度和色度信息融合的分块均值漂移跟踪算法。首先,对跟踪框进行分块,并提取当前帧各子块特征;其次,利用Bhattacharyya距离计算参考目标区域与候选区域间各子块对应的相似程度,根据其相似程度分配各子块的权值,并通过融合规则对各子块相应特征进行融合构成新子块特征,在此基础上,将子块特征选择串接方式作为最终目标特征;最后,采用均值偏移原理迭代估计最终目标位置信息。实验结果表明,当场景中存在相似颜色干扰的情况下,相对于经典均值漂移算法,其准确度提高了84%左右。
To improve the performance of the mean shift tracking algorithm based on single color feature in similar color interference scenarios,a new block mean shift target tracking algorithm combining gradient in-formation with chromaticity color feature fusion is presented. First,tracking box is blocked,and the current frame of each sub-block characteristics is extracted. Second,Bhattacharyya distance is used to calculate the similarity degree of the corresponding sub-block between reference target region and the candidate target area,according to the similarity degree each sub-block weight is distributed,and through the fusion rules each sub-block corresponding features are fuzed to form new sub-block,on this basis,the sub-block fea-tures are concatenated as the ultimate target features. Finally,the mean shift principle is used to iteratively estimate the eventual goal location information. Experimental results show that,compared with the classical mean shift algorithm,the proposed algorithm improves accuracy about 84% in the presence of similar color interference.
出处
《电讯技术》
北大核心
2015年第9期1019-1024,共6页
Telecommunication Engineering
基金
重庆市科技攻关项目(cstc2012gg-yyjs40010)~~
