Multi-target tracking using sample-based data association for mixed images

Ting Hao Zhang, Hsiao Tzu Chen, Chih Wei Tang

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

1 Scopus citations


The ubiquitous specular reflections arose from glasses seriously degrade accuracy of previous visual trackers. Although there have been few visual tracking schemes developed for mixed images with reflections, none focuses on the issue of multi-target tracking. Thus, this paper proposes a multi-target tracking scheme on mixed images with reflections. In the framework of particle filter, the proposed scheme combines the sample-based joint probabilistic data association filter (SJPDAF) with a single target based tracker that uses co-inference and maximum likelihood for visual cue integration to improve tracking accuracy of multiple targets. The co-inference predicted states are used for measurement validation of the SJPDAF. Experimental results show that the proposed scheme works well compared with the SJPDAF tracker.

Original languageEnglish
Title of host publicationAdvances in Visual Computing - 11th International Symposium, ISVC 2015, Proceedings
EditorsMark Elendt, Richard Boyle, Eric Ragan, Bahram Parvin, Rogerio Feris, Tim McGraw, Ioannis Pavlidis, Regis Kopper, George Bebis, Darko Koracin, Zhao Ye, Gunther Weber
PublisherSpringer Verlag
Number of pages11
ISBN (Print)9783319278568
StatePublished - 2015
Event11th International Symposium on Advances in Visual Computing, ISVC 2015 - Las Vegas, United States
Duration: 14 Dec 201516 Dec 2015

Publication series

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


Conference11th International Symposium on Advances in Visual Computing, ISVC 2015
Country/TerritoryUnited States
CityLas Vegas


Dive into the research topics of 'Multi-target tracking using sample-based data association for mixed images'. Together they form a unique fingerprint.

Cite this