Gait recognition based on shape and motion analysis of silhouette contours

This paper presents a three-phase gait recognition method that analyses the spatio-temporal shape and dynamic motion (STS-DM) characteristics of a human subject's silhouettes to identify the subject in the presence of most of the challenging factors that affect existing gait recognition systems. In phase 1, phase-weighted magnitude spectra of the Fourier descriptor of the silhouette contours at ten phases of a gait period are used to analyse the spatio-temporal changes of the subject's shape. A component-based Fourier descriptor based on anatomical studies of human body is used to achieve robustness against shape variations caused by all common types of small carrying conditions with folded hands, at the subject's back and in upright position. In phase 2, a full-body shape and motion analysis is performed by fitting ellipses to contour segments of ten phases of a gait period and using a histogram matching with Bhattacharyya distance of parameters of the ellipses as dissimilarity scores. In phase 3, dynamic time warping is used to analyse the angular rotation pattern of the subject's leading knee with a consideration of arm-swing over a gait period to achieve identification that is invariant to walking speed, limited clothing variations, hair style changes and shadows under feet. The match scores generated in the three phases are fused using weight-based score-level fusion for robust identification in the presence of missing and distorted frames, and occlusion in the scene. Experimental analyses on various publicly available data sets show that STS-DM outperforms several state-of-the-art gait recognition methods.