摘要： 当被跟踪目标受变形、遮挡、快速和不规则运动等因素的干扰时，基于单一颜色特征的相关滤波器跟踪算法难以实现精准的目标定位。为此，分析基于多通道颜色特征Color Names（CN）的核相关滤波器算法（KCF），结合CN特征与颜色统计特征提出一种改进算法。使用掩模矩阵对CN特征的训练样本进行裁切，以提高真实样本的比例。在此基础上，将CN特征与颜色统计特征用于位置相关滤波器的训练，分别获得目标位置，并对两者进行加权处理，得到最终的目标跟踪结果。实验结果表明，与KCF和benchmark_tracker库中性能较优的算法相比，该算法在目标变形、遮挡等干扰下的跟踪精确度和成功率较高。

关键词: 相关滤波器, 目标跟踪, 掩模矩阵, 特征结合, CN特征, 样本裁切

Abstract: When tracking targets are subject to distortion,occlusion,fast and irregular motion,correlation filter tracking algorithms based on single color feature are difficult to achieve accurate target localization.To this end,the Kernel Correlation Filter algorithm(KCF) based on multi-channel color feature of Color Names(CN) is analyzed,and an improved algorithm is proposed integrating CN features and color statistical features.The training samples of the CN features are cut by a mask matrix to increase the proportion of real samples.On this basis,the CN features and the color statistical features are used to train the position-dependent filter to obtain the target positions respectively.And then the two positions are weighted to obtain the final target tracking result.Experimental results show that,compared with KCF and the algorithms with better performance in benchmark_tracker database,the tracking precision and success rate of the proposed algorithm are higher under the inference of distortion and occlusion.

Key words: correlation filter, target tracking, mask matrix, features combination, Color Names(CN) feature, sample cutting

中图分类号: 

• TP391