Use this URL to cite or link to this record in EThOS:
Title: Model-based approaches for recognising people by the way they walk or run
Author: Yam, Chew-Yean
ISNI:       0000 0001 3574 1528
Awarding Body: University of Southampton
Current Institution: University of Southampton
Date of Award: 2002
Availability of Full Text:
Access from EThOS:
Access from Institution:
Using biological traits, such as fingerprints, iris patterns and voice print, in identification and authentication has gained increasing attention due to the demand for a more secure environment. The potential of human walking as a biometric has only attracted interest in the computer vision community since the last decade. Nevertheless, the potential of human running gait as a biometric remains largely unexplored. Here, we propose an approach for an automated non-invasive/markerless person identification system by not only the walking, but also the running gait to explore the potential of these two biomechanically distinct gaits. Two motion models both invariant to walking and running, have been developed based on the concept of harmonic motion. The first is a bilateral symmetric model made up of an upper and a lower pendulum, representing the thigh and the lower leg, joined at the knee. The upper pendulum is simple harmonic motion whilst the lower pendulum uses an empirical model requiring parameter selection for the different gait mode and lacks analytical attributes. The second model has a forced coupled oscillator to describe the knee rotation as legs are considered to be imperfect pendula with energy loss. The rhythm and pattern of gaits are automatically extracted by a temporal evidence gathering technique with the motion models as the underlying temporal templates. The spatio-temporal characteristics of the gait patterns are described by a Fourier representation, which are in turn used to create unique gait signatures for the purpose of identification. Performance analysis demonstrates the potential of gait as a biometric, with running being more potent. This technique not only performs well in discriminating individuals, but also appears capable of distinguishing the gender and gait mode. Moreover, analysis shows that the knee rotation contributes significantly to discrimination capability. Based on the hypothesis that human walking and running gaits are intimately related by the musculo-skeletal structure and that the walking pattern is the phase-modulated version of running (or vice versa), a unique mapping/transform between individuals’ walking and running gait is developed, making the signature invariant to gait mode. Furthermore, this mapping can be used alone as a compressed signature or to buttress the original signature to further improve the recognition capability. Then, a generic relationship between walking and running has been investigated via a neural network. Due to the current size of the experimental dataset, the structure of the two signature spaces could not be drawn, at least not by this approach. However, results do suggest its possible existence. The effect of different camera views is an important application issue. The gait pattern perceived by machine vision at different viewpoints has been investigated. The frequency description of the gait pattern is linearly dependent on the camera sagittal view angle. The changes of both the magnitude and the phase component are symmetric about the fronto-parallel view. This linearity offers a convenient way to map the angular motion obtained from various camera sagittal views to the true motion, for the convenience of gait analysis. More importantly, this linearity can be exploited to develop view invariant gait signatures. The new and interesting findings of this work not only benefit biometrics research, but may also draw attention from other communities such as biomechanics and graphics applications.
Supervisor: Nixon, Mark ; Carter, John Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: QA75 Electronic computers. Computer science ; QH301 Biology ; QM Human anatomy