Discriminative Gait Features Based on Signal Properties of Silhouette Centroids

Abstract

Among the biometric recognition systems, gait recognition plays an important role due to its attractive advantages over other biometric systems. One of the crucial tasks in gait recognition research is the extraction of discriminative features. In this paper, a novel and efficient discriminative feature vector using the signal characteristics of motion of centroids across video frames is proposed. These centroid based features are obtained from the upper and lower regions of the gait silhouette frames in a gait cycle. Since gait cycle contains the sequence of motion pattern and this pattern possesses uniqueness over individuals, extracting the centroid features can better represent the dynamic variations. These variations can be viewed as a signal and therefore the signal properties obtained from the centroid features contains more discriminant information of an individual. Experiments are carried out with CASIA gait dataset B and the proposed feature achieves 97.3% of accuracy using SVM classifier.