Gait Extraction and Description by Evidence-Gathering

Abstract

Using gait as a biometric is of increasing interest, yet there are few model-based, parametric, approaches to extract and describe moving articulated objects. One new approach can detect moving parametric objects by evidence gathering, hence accruing known performance advantages in terms of performance and occlusion. Here we show how that the new technique can be extended not only to extract a moving person, but also to extract and concurrently provide a gait signature for use as a biometric. We show the natural relationaship between the bases of these approaches, and the results they can provide. As such, these techniques allow for gait extraction and description for recognition purposes, and with known performance advantages of a well-established vision technique.