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

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

研究成果: Chapter in Book/Report/Conference proceedingConference contribution

抄録

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.

本文言語英語
ホスト出版物のタイトルAdvances in Swarm Intelligence - 12th International Conference, ICSI 2021, Proceedings
編集者Ying Tan, Yuhui Shi
出版社Springer Science and Business Media Deutschland GmbH
ページ391-406
ページ数16
ISBN(印刷版)9783030788100
DOI
出版ステータス出版済み - 2021
イベント12th International Conference on Advances in Swarm Intelligence, ICSI 2021 - Virtual, Online
継続期間: 7 17 20217 21 2021

出版物シリーズ

名前Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
12690 LNCS
ISSN(印刷版)0302-9743
ISSN(電子版)1611-3349

会議

会議12th International Conference on Advances in Swarm Intelligence, ICSI 2021
CityVirtual, Online
Period7/17/217/21/21

All Science Journal Classification (ASJC) codes

  • 理論的コンピュータサイエンス
  • コンピュータ サイエンス(全般)

フィンガープリント

「Mining Neighbor Frames for Person Re-identification by Global Optimal Tracking」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル