Analysis of Spatio-Temporal Dynamic Patterns of Gait for Recognition

Abstract

This work presents a five-phase automatic gait recognition method that analyzes the spatiotemporal shape and dynamic motion (STS-DM) characteristics of a human subject's silhouette to identify the subject in the presence of many challenging factors that affect gait. Phase-1 describes Krawtchouk Moments for feature extraction; phase-2 describes phase weighted magnitude spectra of the Fourier descriptor of a silhouette. Phase-3 gives a full body shape and motion analysis using ellipses. In Phase-4, dynamic time warping is used to analyze thigh angle rotation pattern. In phase-5 DHT based height varying signal pattern is analyzed. Five phases are fused to give a robust identification system.