一种改进的多通道特征视觉跟踪算法  

An improved multi-channel feature visual tracking algorithm

作　　者：陈建伟[1] 贺继林[1,2] CHEN Jianwei;HE Jilin(State Key Laboratory of High Performance Complex Manufacturing,Central South University,Changsha 410006,China;Sunward Intelligent Equipment Co Ltd,Changsha 410100,China)

机构地区：[1]中南大学高性能复杂制造国家重点实验室,湖南长沙410006 [2]山河智能装备股份有限公司,湖南长沙410100

出　　处：《传感器与微系统》2020年第3期120-123,共4页Transducer and Microsystem Technologies

基　　金：湖南省科技计划项目(2016GK2032);湖南省战略性新兴产业科技攻关项目(2016GK4007)。

摘　　要：为了解决核相关滤波视觉跟踪算法(KCF)因背景干扰和多峰响应导致出现跟踪失败的问题,提出一种多通道融合特征视觉跟踪算法。将目标的颜色通道(CN)特征和方向梯度直方图(HOG)特征融合为新特征,结合主成分分析(PCA)法,保证了与原始的KCF特征对齐。为了解决出现多峰响应干扰的问题,提出一种全新的多峰检测的方法,避免模型跟踪出现漂移。根据OTB50实验的结果,相比其他算法,本文算法具备更好的准确性和鲁棒性,能够提高目标跟踪算法的性能。In order to solve the problem of tracking failure caused by background interference and multi-peak response of kernel correlation filtering(KCF)visual tracking algorithm,a multi-channel fusion feature visual tracking algorithm is proposed.The algorithm fuses the color channel feature(CN)and the direction gradient histogram(HOG)feature of the target into new features,and combines the principal component analysis(PCA)method to ensure alignment with the original KCF features.In order to solve the problem of multi-peak response interference,a new multi-peak detection method is proposed to avoid drift of model tracking.According to the results of the OTB50 experiment,the proposed algorithm has better accuracy and robustness than other algorithms,and can improve the performance of the target tracking algorithm.

关 键 词：核相关滤波 多特征融合 主成分分析法 多峰检测 视觉跟踪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]