Gait recognition based on DWT and SVM

An appearance-based approach to gait recognition is proposed in this paper. The vector data scanned in horizontal, vertical and diagonal direction to the binarized silhouette of a walking person are chosen as the basic gait features. On the basis of the discrete wavelet transformation (DWT), these time spatial feature sequences are decomposed to reduce data dimensionalities and to filter the noise produced from the procedure of template extracting. The multi-class support vector machine (SVM) models are trained by the decomposed feature vectors, and the gaits are classified by the trained SVM models at last. This method is applied to a 30 individual datasets. Extensive experimental results based on NN, KNN and SVM classifier demonstrate that the proposed algorithm would perform an encouraging recognition rate.