单目标跟踪研究综述  

Survey of Single Target Tracking Research

作　　者：闵志方[1] 杜虎 朱雪琼 朱翊翔 王翔[1] MIN Zhi-fang;DU Hu;ZHU Xue-qiong;ZHU Yi-xiang;WANG Xiang(Huazhong Institue of Electro-Optics-Wuhan National Laboratory for Optoelectronics,Wuhan 430223,China;The First Squadron of Naval Maritime Defence and Lifesaving,Qingdao 266000,China;Wuhan Institute of Design and Science,School of Information Engineering,Wuhan 430225,China)

机构地区：[1]华中光电技术研究所—武汉光电国家研究中心,湖北武汉430223 [2]海军海上防险救生第一支队,山东青岛266000 [3]武汉设计工程学院信息工程学院,湖北武汉430225

出　　处：《光学与光电技术》2023年第4期1-14,共14页Optics & Optoelectronic Technology

摘　　要：单目标跟踪作为一项关键的计算机视觉任务在工业与军工领域具有重要作用,从以图像特征为核心的传统方法发展到以网络结构设计为中心的深度学习方法,展现出了较大的研究价值。对单目标跟踪的发展过程进行了总结:首先介绍了一些主流的数据集;接着将以速度为优势的相关滤波方法和以精度高为特色的深度学习类方法作为主要脉络,对其中一些基准方法与高性能方法的设计思路进行了研究分析;最后对各种结构的方法进行了总结,并对今后的研究趋势作出展望。As a challenging visual task,single target tracking plays an important role in the industrial and military fields.From the traditional method with image features as the core to the deep learning method centered on network architecture design,it has shown a lot of research value.The development process of single target tracking is summarized in this paper.Some popular data sets are introduced.The correlation filter methods based on speed and the deep learning methods based on high accuracy are taken as the main principles.The design ideas of some baseline methods and high performance methods are researched.Finally,the methods of various architectures are summarized,and the subsequent research trends are forecasted.

关 键 词：单目标跟踪 深度学习 相关滤波 孪生网络 目标跟踪发展 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]