Gait Recognition Using Wavelet Packet Silhouette Representation and Transductive Support Vector Machines

Abstract

Gait is an idiosyncratic biometric that can be used for human identification at a distance and as a result gained growing interest in intelligent visual surveillance. In this paper, an efficient gait recognition method based on describing subject outer body contour deformations using wavelet packets is proposed. With the use of matching pursuit algorithm, k bases of wavelet packet tree that have maximum similarity to the signal are selected and corresponding coefficients are used as features. Finally, transductive support vector machine (TSVM) classification is utilized on computed eigengait space for semi-supervised identification. The proposed method of selecting features which uses a complete orthogonal or near orthogonal basis from a wavelet packet library of bases and investigating the correlational structure of gait features for each individual using TSVM, result in encouraging identification performance.