GSI: efficient spatio-temporal template for human gait recognition

Abstract

Human gait recognition is a challenging task in computer vision community. In order to represent the gait, the most common feature is a gait template. Many efficient templates have been developed recently, however, the effectiveness of the proposed motion models is still under investigation. A novel template feature, named gait salient image (GSI) is introduced in this paper. The main contribution of the proposed GSI is encoding the motion energy of gait into a single template. This idea is being conceptualised by applying appropriate spatio-temporal filter for extracting motion features and averaging it over a gait period. To show how GSI-based feature is being efficient, the proposed template is classified using PCA+LDA. Extensive experiments on popular gait databases reveal an improvement over the available methods in terms of efficiency and accuracy. The value of recognition rate is 58.44% for Rank1 and 76.60% for Rank5 based on the USF database.