基于高斯尺度空间的核相关滤波目标跟踪算法  

Improved kernel correlation filter tracking with Gaussian scale space

作　　者：谭舒昆 刘云鹏[1] 李义翠 TAN Shukun;LIU Yunpeng;LI Yicui(Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang 110016, China;University of Chinese Academy of Sciences, Beijing 100049, China)

机构地区：[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

出　　处：《计算机工程与应用》2017年第1期29-33,141,共6页Computer Engineering and Applications

基　　金：中国科学院国防科技创新重点基金(No.CXJJ-14-Z65)

摘　　要：核相关滤波(KCF)跟踪算法因其计算效率及速度的优势在目标跟踪领域受到了极大关注,但是该算法仍无法实现尺度自适应,针对此问题提出了一种基于高斯尺度空间的解决方法。根据KCF跟踪算法估计目标位置,将目标及其周围的区域作为搜索区域,并与高斯核卷积建立高斯尺度空间。对高斯尺度空间进行双线性插值,得到目标的多尺度估计图像。用平均绝对误差(MAD)作为匹配准则,将模板与图像匹配,从而得到目标的缩放比率。实验结果表明,与CSK算法、KCF算法等相比,所提出的基于高斯尺度空间的KCF在跟踪精确度上有了显著提升。Recently, Kernel Correlation Filter(KCF)has achieved great attention in visual tracking field, which providesexcellent computation performance and high possessing speed. However, how to handle the scale variation is still an openproblem. Focusing on this issue, a method based on Gaussian scale space is proposed. Firstly, this paper uses KCF to estimatethe location of the target, the context region which includes the target and its surrounding background will be theimage to be matched. In order to get the matching image of a Gaussian scale space, image with Gaussian kernel convolutioncan be got. After getting the Gaussian scale space of the image to be matched, then, according to the Gaussian scale spaceimage, it estimates target image under different scales. It combines with the scale parameter of scale space, for each correspondingscale image performing bilinear interpolation operation to change the size to simulate target imaging at differentscales. Finally, matching the template with different size of images with different scales, the paper uses Mean AbsoluteDifference(MAD)as the match criterion. After getting the optimal matching in the image, it ascertains the best zoom ratios,consequently estimates the target size. In the experiments, compare with CSK, KCF, the results demonstrate that theproposed method achieves high improvement in accuracy and is an efficient algorithm.

关 键 词：目标跟踪 核相关滤波 高斯尺度空间 双线性插值 平均绝对误差 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]