A Feasibility Study of View-independent Gait Identification.
Scientific Publication
- Report Number:
- DSTO-TR-2684
- Authors:
- Whatmough, R.
- Issue Date:
- 2012-03
- AR Number:
- AR-015-265
- Classification:
- Unclassified
- Report Type:
- Technical Report
- Division:
- Intelligence, Surveillance and Reconnaissance Division (ISRD)
- Release Authority:
- Chief, Intelligence, Surveillance and Reconnaissance Division
- Task Sponsor:
- DCDS(IWS)
- Task Number:
- 07/114
- File Number:
- 2010/1173996/1
- Pages:
- 26
- References:
- 32
- Terms:
- Biometric identification; Feature extraction; Image sequences; Image classification; Visual surveillance
- URI:
- http://hdl.handle.net/1947/10190
Abstract
This report reviews work on the identification of people in video recordings by their gait, or walking style. It considers the adjustments needed when the subjects walk other than straight across the image and may not offer the convenient side (fronto-parallel) view. It describes a study of this case using video sequences of walking people, and concludes that the available measurements of gait, apart from head height, are of limited use.
Executive Summary
The automatic analysis of gait, or walking style, of people in video imagery is of great interest as a possible means of matching them with people recorded in a database, at another time or by another nearby camera in a security CCTV network. Existing reviews indicate that gait has been studied in imagery for over twenty years. Gait has the advantage that some aspects of it can be studied in distant views where other features such as the face are not sufficiently resolved for identification.; Most human identification is an attempt to find the best match between a single person and many people in a database. In the security situation it is just as likely to be a comparison between two people who are otherwise new.; Gait is easiest to analyse when the person walks straight across the image and is viewed from the side. Different features of gait appear in a front or rear view, and all features may be present but in weaker form when viewed from intermediate angles. Viewing subjects from above head height adds useful information for the analysis of gait.; Other work on gait identification is reviewed, attention being paid to methods that allow for arbitrary points of view. Some of these methods are incomplete for gait analysis, but could form part of the process.; An experimental study using video sequences of people walking is described. The study has shown that the only gait characteristic that is very useful in distinguishing people and measurable from an above-head camera in any direction is the height.; Any further investigation of the use of other gait characteristics will require more extensive test data and more reliable ways of distinguishing walking persons from their backgrounds.