Human identification based on gait recognition for multiple view angles

Abstract

Gait has emerged as a new biometric verification method which helps in recognising a person by his walking style. In this paper, gait features are extracted based on information set theory, which itself is derived from fuzzy set theory. The uncertainty in the information source values is taken into account by entropy function, based on which gait information image (GII) is derived from a gait cycle. For this purpose a new GII based feature named bipolar sigmoid feature (GII-BPSF) is proposed. Moreover, to address the problem of orientation normalization for different view angles, a modified pre-processing method is adapted from the study of He et al. (The role of size normalization on the recognition rate of handwritten numerals, 2005) to verify the robustness of the proposed features, experiments were carried out on CASIA (Institute of Automation, Chinese Academy of Sciences) dataset B with a wide range of subject variation, different clothing patterns, and carrying conditions. The experimental results show that the proposed GII-BPSF is a more efficient gait representation and feature for an individual recognition and the obtained identification rates are higher concerning the previously established gait recognition approaches.