摘要
行人重识别是计算机视觉领域近年来非常热的一个研究课题,可以被视为图像检索的一个子问题,其目标是给定一个监控行人图像检索跨设备下的该行人图像.传统的方法依赖手工特征,不能适应数据量很大的复杂环境.近年来随着深度学习的发展,大量基于深度学习的行人重识别方法被提出.本文先简单介绍了该问题的定义及传统方法的局限,并列举了一些适用于深度学习方法的行人重识别数据集.此外我们详细地总结了一些比较典型的基于深度学习的行人重识别方法,并比较了部分算法在Market1501数据集上的性能表现.最后我们对该问题未来的研究方向做了一个展望.
Person re-identification(ReID) is a popular research topic in computer vision. It aims to retrieve the given pedestrian image across the device, which can be regarded as a sub-problem of image retrieval. The traditional methods rely on hand-crafted features and can not adapt to the complicated environment with a large number of data. In recent years, with the development of deep learning, a large number of ReID methods based on deep learning have been proposed.This paper briefly introduces the definition of the problem and the limitations of the traditional methods, and then lists some popular databases suitable for deep learning. Moreover, we summarize some typical deep learning based methods in detail, and compare the performance of some algorithms on Market1501. Finally, we make a prospect for the future research direction of person ReID.
作者
罗浩
姜伟
范星
张思朋
LUO Hao;JIANG Wei;FAN Xing;ZHANG Si-Peng(Institute of Cyber-Systems and Control,Zhejiang University,Hangzhou 310027)
出处
《自动化学报》
EI
CSCD
北大核心
2019年第11期2032-2049,共18页
Acta Automatica Sinica
基金
国家自然科学基金(61375049,61633019)
浙江省基础公益研究计划项目(LGF18F030002)资助~~
关键词
行人重识别
深度学习
计算机视觉
卷积神经网络
Person re-identification
deep learning
computer vision
convolutional neural networks
