核相关滤波与孪生网络相结合的目标跟踪算法  

Traget Tracking Algorithm Combining Kernel Correlation Filter and Siamese Network

作　　者：徐亮 张江 张晶[1,2,3] 杨亚琦 XU Liang;ZHANG Jiang;ZHANG Jing;YANG Ya-qi(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China;Yunnan Xiaorun Technology Service Co.,Ltd.,Kunming 650500,China;Yunnan Key Laboratory of Artifical Intelligence,Kunming University of Science and Technology,Kunming 650500,China;Kunming Branch of the 705th Research Institute of China State ShipBuilding Co.,Ltd,Kunming 650102,China;Yunnan Administration for Market Regulation,Kunming 650228,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明650500 [2]云南枭润科技服务有限公司,昆明650500 [3]昆明理工大学云南省人工智能重点实验室,昆明650500 [4]中国船舶集团有限公司第七〇五研究所昆明分部,昆明650102 [5]云南省市场监督管理局,昆明650228

出　　处：《小型微型计算机系统》2021年第4期829-834,共6页Journal of Chinese Computer Systems

基　　金：云南省技术创新人才项目(2019HB113)资助;云南省“万人计划”产业技术领军人才项目(云发改人事[2019]1096号)资助。

摘　　要：针对核相关滤波目标跟踪算法中对局部上下文区域图像提取的HOG特征图在复杂环境下不能保证目标跟踪的精度问题,提出了一种核相关滤波与孪生网络相结合的目标跟踪算法.首先在首帧输入图像中提取HOG特征图并建立相关滤波器模板,同时提取经过孪生网络的目标区域图像特征图;然后若后续帧输入图像帧数不为5的倍数则提取仿射变换HOG特征图,否则提取经过孪生网络的搜索区域图像特征图;最后根据遮挡处理的结果自适应获取目标位置并更新模型和最终相关滤波器模板.仿真实验结果表明本文算法在保证目标跟踪精度的前提下具有满足实时跟踪要求的跟踪速率.Aiming at the problem that the HOG feature map extracted from the local context area image in the kernel correlation filter target tracking algorithm can’t guarantee the accuracy of traget tracking in complex environments.A target tracking algorithm combining kernel correlation filter and siamese network is proposed.Firstly,extracting the HOG feature map to establish a correlation filter template in the first frame input image,and simultaneously extracting target area image feature map through siamese network.Then if the number of the input image frame in subsequent frames is not a multiple of five,extracting the affine transformation HOG feature map,otherwise extracting serach area image feature map through siamese network.Finally,according to the result of occlusion process to adaptively obtain the target position and update model and the final correlation filter template.Simulation results show that the algorithm in this paper has a tracking rate that meets the requirements of real-time tracking on the premise of ensuring traget tracking accuracy.

关 键 词：核相关滤波 目标跟踪 孪生网络 特征图更新 模板更新 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]