Biologically-inspired motion detection and classification : human and machine perception
Biologically-inspired motion detection and classification : human and machine perception
Humans are good at the perception of biological motion, i.e. the motion of living things. The human perceptual system can tolerate not only variations in lighting conditions, distance, etc., but it can also perceive such motion and categorise it as walking, running, jumping etc. from minimal information systems such as moving light displays (MLDs). In these displays only specific points (e.g. joints in the case of a human being) are visible. Although a static display looks like a random configuration of dots, a dynamic display is perceptually organised into a moving figure. Some kind of temporal integration of the spatial contents seems to be a part of the perception mechanism; as manifested from the minimum presentation time required for biological motion to become apparent. One possible way to understand human perception may be to build an equivalent machine model. An analysis of the workings of this machine may lend us an insight into human perception. In this work, we considered a closed set of 12 different categories of MLD sequences. These sequences were shown to 93 participants and their responses are used as the basis of comparison of human and machine perception. Human responses were compared with the performance of /c-nearest neighbour and neural network detectors. Machine perception is found to differ from human perception in some important respects. We also examined the related aspect of person identification on the basis of gait. This has important applications in the fields of surveillance and biometrics. In recent years, gait has been investigated as a potential biometric; as this may be the only information available to identify a distant and/or otherwise masked person. Humans can learn to recognise different subjects in MLDs. In our experiments with a dataset of 21 subjects, an accuracy of nearly 90% and 100% was achieved with neural network and support vector machine classifiers respectively. Also the machines were able to make this recognition in a fraction of a gait cycle.
University of Southampton
Laxmi, Vijay
e79a9a3a-cd6b-45d6-8a03-9a5891453a95
2003
Laxmi, Vijay
e79a9a3a-cd6b-45d6-8a03-9a5891453a95
Carter, John N
e05be2f9-991d-4476-bb50-ae91606389da
Damper, Robert
6e0e7fdc-57ec-44d4-bc0f-029d17ba441d
Laxmi, Vijay
(2003)
Biologically-inspired motion detection and classification : human and machine perception.
University of Southampton, Electronics and Computer Science : Faculty of Engineering and Applied Science, Doctoral Thesis.
Record type:
Thesis
(Doctoral)
Abstract
Humans are good at the perception of biological motion, i.e. the motion of living things. The human perceptual system can tolerate not only variations in lighting conditions, distance, etc., but it can also perceive such motion and categorise it as walking, running, jumping etc. from minimal information systems such as moving light displays (MLDs). In these displays only specific points (e.g. joints in the case of a human being) are visible. Although a static display looks like a random configuration of dots, a dynamic display is perceptually organised into a moving figure. Some kind of temporal integration of the spatial contents seems to be a part of the perception mechanism; as manifested from the minimum presentation time required for biological motion to become apparent. One possible way to understand human perception may be to build an equivalent machine model. An analysis of the workings of this machine may lend us an insight into human perception. In this work, we considered a closed set of 12 different categories of MLD sequences. These sequences were shown to 93 participants and their responses are used as the basis of comparison of human and machine perception. Human responses were compared with the performance of /c-nearest neighbour and neural network detectors. Machine perception is found to differ from human perception in some important respects. We also examined the related aspect of person identification on the basis of gait. This has important applications in the fields of surveillance and biometrics. In recent years, gait has been investigated as a potential biometric; as this may be the only information available to identify a distant and/or otherwise masked person. Humans can learn to recognise different subjects in MLDs. In our experiments with a dataset of 21 subjects, an accuracy of nearly 90% and 100% was achieved with neural network and support vector machine classifiers respectively. Also the machines were able to make this recognition in a fraction of a gait cycle.
Text
996903.pdf
- Version of Record
Text
VL_Thesis.pdf
- Version of Record
More information
Published date: 2003
Identifiers
Local EPrints ID: 257383
URI: http://eprints.soton.ac.uk/id/eprint/257383
PURE UUID: 1c014dce-fde7-46cf-85a3-57917bc6109d
Catalogue record
Date deposited: 04 Oct 2004
Last modified: 14 Mar 2024 05:56
Export record
Contributors
Author:
Vijay Laxmi
Thesis advisor:
John N Carter
Thesis advisor:
Robert Damper
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