Mining Neighbor Frames for Person Re-identification by Global Optimal Tracking

Kai Han, Jinho Lee, Lang Huang, Fangcheng Liu, Seiichi Uchida, Chao Zhang

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Person re-identification is a challenging task aiming to identify the same person across different cameras. However, most of existing image-based person re-identification methods neglect the spatial and temporal constraint, the information in neighbor frames of each person image is rarely exploited by previous studies. In this paper, we propose a novel neighbor frames mining framework (NFM) to exploit the spatial-temporal information. For each gallery image, we use a dynamic programming-based global optimal tracking method to search images of the same person in its neighbor frames. From those images, the image features extracted by the shared convolutional neural network (CNN) in the constructed neighbor sequence are merged via an attention weighted averaging technology. To this end, a novel supervised attention mechanism is designed for dealing with tracking errors. The final feature with multi-view and robust information is used for matching. Experimental results show the superiority and efficiency of the proposed method on two benchmark datasets including DukeMTMC-reID and PRW.

Original languageEnglish
Title of host publicationAdvances in Swarm Intelligence - 12th International Conference, ICSI 2021, Proceedings
EditorsYing Tan, Yuhui Shi
PublisherSpringer Science and Business Media Deutschland GmbH
Pages391-406
Number of pages16
ISBN (Print)9783030788100
DOIs
Publication statusPublished - 2021
Event12th International Conference on Advances in Swarm Intelligence, ICSI 2021 - Virtual, Online
Duration: Jul 17 2021Jul 21 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12690 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference12th International Conference on Advances in Swarm Intelligence, ICSI 2021
CityVirtual, Online
Period7/17/217/21/21

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint

Dive into the research topics of 'Mining Neighbor Frames for Person Re-identification by Global Optimal Tracking'. Together they form a unique fingerprint.

Cite this