面向行人重识别分集特征挖掘的多分支深度神经网络构造进展  

Recent advances on construction of multi⁃branch neural networks for person re⁃identification

作　　者：吴晓富[1] 尹梓睿 宋越 张磊[1] 谢奔 赵师亮 张索非[3] WU Xiaofu;YIN Zirui;SONG Yue;ZHANG Lei;XIE Ben;ZHAO Shiliang;ZHANG Suofei(College of Telecommunications&Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;Unit 95958 of PLA,Shanghai 200120,China;School of Internet of Things,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003 [2]95958部队,上海200120 [3]南京邮电大学物联网学院,江苏南京210003

出　　处：《南京邮电大学学报（自然科学版）》2021年第1期78-85,共8页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

基　　金：国家自然科学基金(61671253)资助项目。

摘　　要：随着深度学习的兴起,近年来行人重识别性能提高的关键是如何构建具有分集特征挖掘的深度神经网络以获得行人图像的紧凑表达,而多分支深度神经网络是实现分集特征挖掘的主要途径。文中分析了近年来用于行人重识别多分支深度神经网络的几种主流构建方法,包括局部分割分集、注意力分集、特征遮挡分集、异构支路分集等,并在3个行人重识别数据集Market1501,Duke,CU⁃HK03上对不同算法性能进行了比较分析。基于当前进展,最后还讨论了行人重识别分集特征挖掘的未来研究趋势。With the surge of deep learning,there is increased interest on the design of multi⁃branch neural networks for extracting rich features in person re⁃identification.This paper reviews some recent ad⁃vances on this topic,including part⁃based approaches,attention⁃based approaches,dropping approaches,and heterogeneous approaches.Various state⁃of⁃the⁃art approaches on the popular pedestrian re⁃identification datasets,including Market1501,Duke and CUHK03,are compared and analyzed,re⁃spectively.Finally,future directions of the person re⁃identification with diverse features are discussed.

关 键 词：行人重识别 多分支网络 分集特征 特征嵌入 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]