Das Choudhury, Sruti (2013) Uniscale and multiscale gait recognition in realistic scenario. PhD thesis, University of Warwick.

Preview

Text WRAP_THESIS_Das_Choudhury_2013.pdf - Submitted Version Download (7Mb) | Preview

Request Changes to record.

Abstract

The performance of a gait recognition method is affected by numerous challenging
factors that degrade its reliability as a behavioural biometrics for subject identification in
realistic scenario. Thus for effective visual surveillance, this thesis presents five gait recog-
nition methods that address various challenging factors to reliably identify a subject in
realistic scenario with low computational complexity. It presents a gait recognition method
that analyses spatio-temporal motion of a subject with statistical and physical parameters
using Procrustes shape analysis and elliptic Fourier descriptors (EFD). It introduces a part-
based EFD analysis to achieve invariance to carrying conditions, and the use of physical
parameters enables it to achieve invariance to across-day gait variation. Although spatio-
temporal deformation of a subject’s shape in gait sequences provides better discriminative
power than its kinematics, inclusion of dynamical motion characteristics improves the iden-
tification rate. Therefore, the thesis presents a gait recognition method which combines
spatio-temporal shape and dynamic motion characteristics of a subject to achieve robust-
ness against the maximum number of challenging factors compared to related state-of-the-
art methods. A region-based gait recognition method that analyses a subject’s shape in
image and feature spaces is presented to achieve invariance to clothing variation and carry-
ing conditions. To take into account of arbitrary moving directions of a subject in realistic
scenario, a gait recognition method must be robust against variation in view. Hence, the the-
sis presents a robust view-invariant multiscale gait recognition method. Finally, the thesis
proposes a gait recognition method based on low spatial and low temporal resolution video
sequences captured by a CCTV. The computational complexity of each method is analysed.
Experimental analyses on public datasets demonstrate the efficacy of the proposed methods.

Item Type:	Thesis (PhD)
Subjects:	Q Science > QH Natural history > QH301 Biology T Technology > TA Engineering (General). Civil engineering (General)
Library of Congress Subject Headings (LCSH):	Biometry, Gait in humans -- Mathematical models, Pattern recognition systems, Fourier analysis, Computer vision
Official Date:	October 2013
Institution:	University of Warwick
Theses Department:	School of Engineering
Thesis Type:	PhD
Publication Status:	Unpublished
Supervisor(s)/Advisor:	Tjahjadi, Tardi
Sponsors:	Warwick Postgraduate Research Scholarship (WPRS); University of Warwick. Department of Engineering
Extent:	xx,163 leaves : charts.
Language:	eng

Request changes or add full text files to a record

Repository staff actions (login required)

View Item

Downloads