Computer Engineering and Applications ›› 2018, Vol. 54 ›› Issue (9): 178-182.DOI: 10.3778/j.issn.1002-8331.1612-0316

Previous Articles     Next Articles

Improved kernelized correlation filter tracking

SUN Jian1,2, XIANG Wei1, TAN Shukun1, 2, LIU Yunpeng1   

  1. 1.Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang 110016, China
    2.University of Chinese Academy of Sciences, Beijing 100049, China
  • Online:2018-05-01 Published:2018-05-15


孙  健1,2,向  伟1,谭舒昆1,2,刘云鹏1   

  1. 1.中国科学院 沈阳自动化研究所,沈阳 110016
    2.中国科学院大学,北京 100049

Abstract: An improved kernel correlation filtering target tracking algorithm is proposed by Kernelized Correlation Filtering(KCF) tracking algorithm, which can not solve the problem of scale change and out-of-view in target tracking. Firstly, a scale filter is added to improve the target scale change based on training translation filter. In order to solve out-of-view problem, the occlusion processing mechanism is used. When the target is not completely occluded, the SVM is used to train the samples, and the re-detection classifier is adopted to detect. Experimental results show that the tracking accuracy of this method is obviously improved in comparison with other excellent tracking algorithms.

Key words: kernel correlation filter, out-of-view, scale change, occlusion, support vector machine

摘要: 主要针对核相关滤波(KCF)跟踪算法无法解决目标跟踪中尺度变化及目标丢失问题,提出了一种改进的核相关滤波目标跟踪算法。在训练位移滤波器的基础上增加了一个尺度滤波器来改进目标尺度变化问题。为解决目标丢失问题,结合了遮挡处理机制,当判断目标受到遮挡面积较小时使用支持向量机(SVM)对样本进行在线训练,当目标遮挡时使用再检测分类器进行检测。实验结果表明,该方法与其他优秀跟踪算法比较跟踪精度有明显提升。

关键词: 核相关滤波, 目标丢失, 尺度变化, 遮挡, 支持向量机