Recognizing human gait by model-driven statistical analysis
Recognizing human gait by model-driven statistical analysis
The ability to recognize humans by computer vision is a very important task, with many potential applications. In this thesis, we present a new method for an automated marker-less system to describe, analyze and recognize the human gait motion. The automated system consists of four stages: i) detection and extraction of the moving human body and its contour from image sequences; ii) extraction of human gait signatures based on topological analysis guided by known anatomical knowledge; iii) description of gait parameters by statistical analysis of the gait signatures; and iv) feature extraction and recognition of human gait. The gait signature is represented by a sequential set of 2D stick figures during one gait cycle. A grammatical structure with constraints of the gait sequences has been developed to improve the robustness of the gait signature, together with a new method of step symmetry. In the gait signature, the motion parameters based on biomechanical studies are calculated for characterizing the human gait. The inherent periodicity in gait motion is detected by graphical methods and analyzed by statistical approaches. Also, the periodic gait modern is modelled by interpolation of trigonometric-polynomials. In addition, the features based on motion parameters are extracted from the sequence of gait signatures. Then, a k-nearest neighbour classifier and an enhanced back-propagation algorithm is employed to recognize the gait. In experiments, the proposed methods have been successfully demonstrated on the largest available database. The gait signature is a very effective and well-defined representation method for analyzing the gait motion. It can be applied to other areas such as biomechanical and clinical applications.
University of Southampton
Yoo, Jang-Hee
cbfc2f5d-2a17-4f6d-a94c-788b5747fa7b
2004
Yoo, Jang-Hee
cbfc2f5d-2a17-4f6d-a94c-788b5747fa7b
Yoo, Jang-Hee
(2004)
Recognizing human gait by model-driven statistical analysis.
University of Southampton, Doctoral Thesis.
Record type:
Thesis
(Doctoral)
Abstract
The ability to recognize humans by computer vision is a very important task, with many potential applications. In this thesis, we present a new method for an automated marker-less system to describe, analyze and recognize the human gait motion. The automated system consists of four stages: i) detection and extraction of the moving human body and its contour from image sequences; ii) extraction of human gait signatures based on topological analysis guided by known anatomical knowledge; iii) description of gait parameters by statistical analysis of the gait signatures; and iv) feature extraction and recognition of human gait. The gait signature is represented by a sequential set of 2D stick figures during one gait cycle. A grammatical structure with constraints of the gait sequences has been developed to improve the robustness of the gait signature, together with a new method of step symmetry. In the gait signature, the motion parameters based on biomechanical studies are calculated for characterizing the human gait. The inherent periodicity in gait motion is detected by graphical methods and analyzed by statistical approaches. Also, the periodic gait modern is modelled by interpolation of trigonometric-polynomials. In addition, the features based on motion parameters are extracted from the sequence of gait signatures. Then, a k-nearest neighbour classifier and an enhanced back-propagation algorithm is employed to recognize the gait. In experiments, the proposed methods have been successfully demonstrated on the largest available database. The gait signature is a very effective and well-defined representation method for analyzing the gait motion. It can be applied to other areas such as biomechanical and clinical applications.
Text
973833.pdf
- Version of Record
More information
Published date: 2004
Identifiers
Local EPrints ID: 465567
URI: http://eprints.soton.ac.uk/id/eprint/465567
PURE UUID: d2e1a1a0-3a43-4f9b-a3be-7e737b25c9bb
Catalogue record
Date deposited: 05 Jul 2022 01:50
Last modified: 16 Mar 2024 20:15
Export record
Contributors
Author:
Jang-Hee Yoo
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics