A Dual Scale Matching Model for Long-Term Association

Zhen Ye, Yubin Wu, Shuai Wang, Yang Zhang, Yanbing Chen, Wei Ke, Hao Sheng

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Citation (Scopus)


Multi-object tracking can be characterized as a data association problem. The advantage of RNN in processing temporal dependence makes it an ideal selection in data association. When factors such as scene congestion and weak illumination cause detection failure especially long intervals, association is often very difficult and eventually leads to tracking failure. To solve this problem, Dual Scale Matching model (DSM) containing a Motion Trend Match Network (MTMNet) and an Appearance History Memory Network (AHMNet) is proposed. DSM is a long-term optimization method based on multiple hypothesis tracking. MTMNet aims to learn a similarity metric matching function between tracklets leveraging the motion feature. AHMNet is designed to provide optimal pruning strategies leveraging long period appearance feature. Our method is effective on MOT17 benchmark and it shows that we achieve considerable competitive results with current state-of-the-art trackers.

Original languageEnglish
Title of host publicationWireless Algorithms, Systems, and Applications - 15th International Conference, WASA 2020, Proceedings
EditorsDongxiao Yu, Falko Dressler, Jiguo Yu
PublisherSpringer Science and Business Media Deutschland GmbH
Number of pages13
ISBN (Print)9783030590154
Publication statusPublished - 2020
Event15th International Conference on Wireless Algorithms, Systems, and Applications, WASA 2020 - Qingdao, China
Duration: 13 Sept 202015 Sept 2020

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12384 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Conference15th International Conference on Wireless Algorithms, Systems, and Applications, WASA 2020


  • Appearance History Memory Net
  • Data association
  • Dual Scale Matching
  • Motion Trend Match Net
  • Multi-object tracking


Dive into the research topics of 'A Dual Scale Matching Model for Long-Term Association'. Together they form a unique fingerprint.

Cite this