结合目标不变矩的核相关滤波跟踪算法  

Kernelized correlation filters algorithm combined with target invariant moments

作　　者：王凯[1] 黄山[1] WANG Kai;HUANG Shan(College of Electrical Engineering,Sichuan University,Chengdu 610065,China)

机构地区：[1]四川大学电气工程学院,四川成都610065

出　　处：《计算机工程与设计》2020年第8期2271-2277,共7页Computer Engineering and Design

摘　　要：为提升核相关滤波跟踪算法(KCF)在目标出现遮挡时的鲁棒性,提出一种结合不变矩特征的核相关滤波跟踪算法。以不变矩特征描述目标,通过初始模型与待测目标之间的相似度变化情况,设定遮挡判断机制;利用相似度的大小将模型更新机制中的学习率分段,实现目标模型的自适应更新。为测试算法的有效性,采用OTB-2013评估数据集,实验结果表明,与KCF算法相比,该算法在跟踪精度上提升了7.4%,在成功率上提升了10.8%。To improve the robustness of the kernelized correlation filters tracking algorithm(KCF)when the target appears occlusion,a kernel correlation filtering tracking algorithm combined with invariant moment features was proposed.The invariant moment feature was used to describe the target,and the occlusion judgment mechanism was established through the similarity change between the initial model and the target to be tested.In this way,the learning rate in the model update mechanism was divided into different values using the magnitude of similarity,so that the adaptive updating of the target model was realized.OTB-2013 data set was used to evaluate the effectiveness of the algorithm.Experimental results show that,compared with KCF algorithm,the proposed algorithm improves tracking accuracy by 7.4%and success rate by 10.8%.

关 键 词：目标跟踪 遮挡 核相关滤波 不变矩特征 学习率 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]