Video tracking of people under severe occlusions

Zhang, Z

Video tracking of people under severe occlusions

Zhang, Z

Permalink

Publication Type:: Thesis
Issue Date:: 2010

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download contents and abstractAdobe PDF (112.97 kB)

Adobe PDF

Download thesisAdobe PDF (12.26 MB)

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Zhang, Z
dc.date.accessioned	2010-12-06T22:35:31Z
dc.date.accessioned	2012-12-15T03:53:18Z
dc.date.available	2010-12-06T22:35:31Z
dc.date.available	2012-12-15T03:53:18Z
dc.date.issued	2010
dc.identifier.uri	http://hdl.handle.net/10453/20313
dc.description	University of Technology, Sydney. Faculty of Engineering and Information Technology.
dc.description.abstract	Video surveillance in dynamic scenes, especially for humans and vehicles, is currently one of the most active research topics in computer vision and pattern recognition. The goal of this research is to develop a real-time automatic tracking system which is both reliable and efficient by utilizing computational approaches. The literature has presented many valuable methods on object tracking; however, most of those algorithms can only perform effectively under simple scenarios. There are a few algorithms which attempt to accomplish object tracking in a complex dynamic scene and have successfully achieved their goals when the dynamic scene is not too complex. However no system yet is capable of accurately handling object tracking, especially human tracking, in a crowded environment with frequent and continuous occlusions. Therefore, the goal of my research is to develop an effective human tracking algorithm which takes into account and overcomes the various factors involved in a complex dynamic scene. The founding idea is that of dividing the human figure into five main parts, and track each individually under a constraint of integrity. Data association in new frames is performed on each part, and is inferred for the whole human figure through a fusion rule. This approach has proved a good trade off between model complexity and actual computability. Experimental results have confirmed the effectiveness of the methodology.	en
dc.format	Thesis (PhD)	en
dc.language.iso	en	en
dc.relation	https://opus.lib.uts.edu.au/bitstream/10453/20313/2/02Whole.pdf
dc.relation.replaces	http://hdl.handle.net/2100/1201
dc.rights	au.edu.uts.lib/ppc
dc.rights	info:eu-repo/semantics/openAccess
dc.rights	The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
dc.subject	Video surveillance.	en
dc.subject	Computer vision.	en
dc.subject	Pattern recognition systems.	en
dc.title	Video tracking of people under severe occlusions	en
dc.type	Thesis
utslib.copyright.status	open_access

Abstract:

Video surveillance in dynamic scenes, especially for humans and vehicles, is currently one of the most active research topics in computer vision and pattern recognition. The goal of this research is to develop a real-time automatic tracking system which is both reliable and efficient by utilizing computational approaches. The literature has presented many valuable methods on object tracking; however, most of those algorithms can only perform effectively under simple scenarios. There are a few algorithms which attempt to accomplish object tracking in a complex dynamic scene and have successfully achieved their goals when the dynamic scene is not too complex. However no system yet is capable of accurately handling object tracking, especially human tracking, in a crowded environment with frequent and continuous occlusions. Therefore, the goal of my research is to develop an effective human tracking algorithm which takes into account and overcomes the various factors involved in a complex dynamic scene. The founding idea is that of dividing the human figure into five main parts, and track each individually under a constraint of integrity. Data association in new frames is performed on each part, and is inferred for the whole human figure through a fusion rule. This approach has proved a good trade off between model complexity and actual computability. Experimental results have confirmed the effectiveness of the methodology.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/20313