Novel Signatures for Gait Analysis Using Level Sets

Abstract

Gait analysis has gained prominence towards detecting suspicious activity in public area which is generally kept under surveillance. In this paper we present the results of recognizing a person based on his gait movement. Novel signatures were built through distance mapped level set contour based on boundary and skeletal information. Gait video sequence experimented upon is silhouette corresponding to 16 subjects with 50 frames per subject. Results were found to be encouraging at first level where signatures have been compared with available subjects' gait video. The robustness is tested with morphological distortions on the query image. The subject has been found to be detected with reasonable acceptance.