Sparse multilinear Laplacian discriminant analysis for gait recognition

Distinct and unique ways of walking have attracted the attention of researchers for identification of different individuals. The Gait; the walking video sequence offers unique advantages like the capability of capturing from a distance in a non-cooperative environment and providing recognition without being obtrusive. These advantages has demonstrated applications of gait recognition during intelligent video surveillance. The paper presents new discriminative feature extraction and dimensionality reduction scheme called as Sparse Multilinear Laplacian Discriminant Analysis (SMLDA) for gait recognition. SMLDA exploits the benefits of Laplacian weighted scatter difference instead of simple scatter difference and sparsity constraint as a class separability measure. The performance of the proposed scheme has been evaluated experimentally on CASIA, USF and OU-ISIR datasets. The experimental results show the competitive performance in comparison with conventional gait recognition schemes.