工程科学学报.第44卷，第X期：1-13.2021年X月 Chinese Journal of Engineering,Vol.44,No.X:1-13,X 2021 https://doi.org/10.13374/j.issn2095-9389.2020.12.22.004;http://cje.ustb.edu.cn 基于深度学习的行人重识别方法综述 李擎12，，胡伟阳l2)，李江昀，2)四，刘艳，2)，李梦璇2 1)北京科技大学自动化学院，北京1000832)工业过程知识自动化教育部重点实验室，北京1000833)北京科技大学顺德研究生院.佛 山528000 ☒通信作者，E-mail:leejy@ustb.edu.cn 摘要对深度学习在行人重识别领域的应用现状进行总结与评价.首先，对行人重识别进行介绍，包括行人重识别的应用 场景、数据集与评价指标，并对基于深度学习的行人重识别的基本方法进行总结.之后，针对行人重识别的研究现状，将近年 来国内外学者的研究工作归纳为基于局部特征、基于生成对抗网络、基于视频以及基于重排序4个方向，并对每个方向所使 用的方法分别进行梳理、性能对比以及总结.最后，对行人重识别领域现存的问题进行了分析与讨论，并探讨了行人重识别 未来的发展方向 关键词深度学习：行人重识别：局部特征；生成对抗网络；视频数据：重排序 分类号TG183 A survey of person re-identification based on deep learning LI Qing2,HU Wei-yang2,LI Jiang-yun LIU Yan 2,LI Meng-xuan2) 1)School of Automation&Electrical Engineering.University of Science and Technology Beijing.Beijing 100083,China 2)Key Laboratory of Knowledge Automation for Industrial Processes,Ministry of Education,Beijing 100083,China 3)Shunde Graduate School,University of Science and Technology Beijing.Foshan 528000,China Corresponding author,E-mail:leejy@ustb.edu.cn ABSTRACT Person re-identification is an important part of multi-target tracking across cameras;its aim is to identify the same person across different cameras.Given a query image,the purpose of person re-identification is to find the best match for the query image in an image set.Person re-identification is a key component in an intelligent security system;it is beneficial for building a smart bank or smart factory and plays a crucial role in the construction of a smart city.Nowadays,with the development of artificial intelligence and increasing demand for precise identification in practical scenarios,deep learning-based person re-identification technology has become a popular research topic;this technology has achieved state-of-the-art results in comparison with conventional approaches.Although there are many recently proposed networks with stronger representation ability and a high level of accuracy for person re-identification,there also exist some problems that should be considered and solved.These include the insufficient generalization ability of various poses,the inability to fully utilize the temporal information,and the ineffective identification of occluded objects.As a result,many scholars have researched this field and have pointed out some promising solutions to cope with the aforementioned problems.This paper aims to summarize the application of deep learning in the field of person re-identification along with its advantages and shortcomings.First,the background of person re-identification is introduced,including the application scenarios,datasets,and evaluation indicators. Additionally,some basic methods of person re-identification based on deep learning are summarized.According to the existing research on person re-identification,the main approaches proposed by scholars worldwide can be summarized into four aspects,which are based 收稿日期：2020-12-22 基金项目：中央高校基本科研业务费专项资金资助项目(FFDF19-0O2):北京科技大学顺德研究生院科技创新专项资金资助项目 (BK20BE014)基于深度学习的行人重识别方法综述 李    擎1,2,3)，胡伟阳1,2)，李江昀1,2,3) 苣，刘    艳1,2)，李梦璇1,2) 1) 北京科技大学自动化学院，北京 100083    2) 工业过程知识自动化教育部重点实验室，北京 100083    3) 北京科技大学顺德研究生院，佛 山 528000 苣通信作者， E-mail:leejy@ustb.edu.cn 摘    要    对深度学习在行人重识别领域的应用现状进行总结与评价. 首先，对行人重识别进行介绍，包括行人重识别的应用 场景、数据集与评价指标，并对基于深度学习的行人重识别的基本方法进行总结. 之后，针对行人重识别的研究现状，将近年 来国内外学者的研究工作归纳为基于局部特征、基于生成对抗网络、基于视频以及基于重排序 4 个方向，并对每个方向所使 用的方法分别进行梳理、性能对比以及总结. 最后，对行人重识别领域现存的问题进行了分析与讨论，并探讨了行人重识别 未来的发展方向. 关键词    深度学习；行人重识别；局部特征；生成对抗网络；视频数据；重排序 分类号    TG183 A survey of person re-identification based on deep learning LI Qing1,2,3) ，HU Wei-yang1,2) ，LI Jiang-yun1,2,3) 苣 ，LIU Yan1,2) ，LI Meng-xuan1,2) 1) School of Automation & Electrical Engineering, University of Science and Technology Beijing, Beijing 100083, China 2) Key Laboratory of Knowledge Automation for Industrial Processes, Ministry of Education, Beijing 100083, China 3) Shunde Graduate School, University of Science and Technology Beijing, Foshan 528000, China 苣 Corresponding author, E-mail: leejy@ustb.edu.cn ABSTRACT    Person re-identification is an important part of multi-target tracking across cameras; its aim is to identify the same person across different cameras. Given a query image, the purpose of person re-identification is to find the best match for the query image in an image set. Person re-identification is a key component in an intelligent security system; it is beneficial for building a smart bank or smart factory  and  plays  a  crucial  role  in  the  construction  of  a  smart  city.  Nowadays,  with  the  development  of  artificial  intelligence  and increasing demand for precise identification in practical scenarios, deep learning-based person re-identification technology has become a popular research topic; this technology has achieved state-of-the-art results in comparison with conventional approaches. Although there are many recently proposed networks with stronger representation ability and a high level of accuracy for person re-identification, there also exist some problems that should be considered and solved. These include the insufficient generalization ability of various poses, the inability to fully utilize the temporal information, and the ineffective identification of occluded objects. As a result, many scholars have researched  this  field  and  have  pointed  out  some  promising  solutions  to  cope  with  the  aforementioned  problems.  This  paper  aims  to summarize the application of deep learning in the field of person re-identification along with its advantages and shortcomings. First, the background  of  person  re-identification  is  introduced,  including  the  application  scenarios,  datasets,  and  evaluation  indicators. Additionally, some basic methods of person re-identification based on deep learning are summarized. According to the existing research on person re-identification, the main approaches proposed by scholars worldwide can be summarized into four aspects, which are based 收稿日期: 2020−12−22 基金项目: 中央高校基本科研业务费专项资金资助项目（FRF-DF-19-002）；北京科技大学顺德研究生院科技创新专项资金资助项目 （BK20BE014） 工程科学学报，第 44 卷，第 X 期：1−13，2021 年 X 月 Chinese Journal of Engineering, Vol. 44, No. X: 1−13, X 2021 https://doi.org/10.13374/j.issn2095-9389.2020.12.22.004; http://cje.ustb.edu.cn