Gait recognition based on outermost contour

Abstract

Gait recognition aims to identify people by the way they walk. In this paper, a simple but effective gait recognition method based on Outermost Contour is proposed. For each gait image sequence, an adaptive silhouette extraction algorithm is firstly used to segment the images and a series of postprocessing is applied to the silhouette images to obtain the normalized silhouettes with less noise. Then a novel feature extraction method based on Outermost Contour is proposed. Principal Component Analysis (PCA) and Multiple Discriminant Analysis (MDA) are adopted to reduce the dimensionality of the feature vectors and to optimize the class separability of different gait image sequences simultaneously. Two simple pattern classification methods are used on the low-dimensional eigenspace for recognition. Experimental results on a gait database of 100 people show that the accuracy of our algorithm achieves 97.67%. © 2010 Springer-Verlag Berlin Heidelberg.