行人重识别研究综述  

A REVIEW ON PERSON RE-IDENTIFICATION

作　　者：张化祥[1] 刘丽[1] Zhang Huaxiang;Liu Li(School of Information Science and Engineering,Shandong Normal University,250358,Jinan,China)

机构地区：[1]山东师范大学信息科学与工程学院,济南250358

出　　处：《山东师范大学学报（自然科学版）》2018年第4期379-387,共9页Journal of Shandong Normal University(Natural Science)

基　　金：国家自然科学基金资助项目(61572298);国家自然科学基金资助项目(61702310)

摘　　要：行人重识别是指实现不重叠的不同摄像头下同一行人图像的匹配技术,在加强社会管理、预防犯罪行为发生以及实现事件重构等方面具有重要应用价值.由于行人重识别主要依靠人体外表视觉表示特征和人工设计特征,且受光照、图像分辨率、行人姿态及拍摄视角度等因素的影响较大,因此,行人重识别面临巨大挑战.本文对现有行人表示特征学习技术及度量技术进行了综述分析,指出存在的问题及可能的解决思路.本文的论述有利于该领域研究人员对现状的把握及提出新的研究思路.Person Re-identification(ReID)refers to the matching technology of the same person images under different non-overlapping cameras.It can be widely applied in strengthening social management,criminal prevention and event reconstruction.ReID mainly relies on the visual presentations and hand-crafted features of person appearance,which are greatly influenced by illumination condition,image resolution,person posture and shooting angle,so ReID faces great challenges.This paper reviews and analyzes the existing person representation feature learning technologies and measurement technologies,together with the existing issues and possible solutions,which may be helpful to those doing research works in this domain to grasp the research progress and put forward new research ideas.

关 键 词：行人重识别 图像特征 深度学习 语义信息 

分 类 号：TQ3914[化学工程]