Gait Recognition Using Density-Based Outlier Detection and Location Fusion by Sparse Representation

Abstract

A novel algorithm to recognize human identities via accurate gait features by sparse representation is proposed. Acceleration-based gait recognition is a continuous biometric recognize method, which is easy to accept. The proposed algorithm firstly judges abnormal signature point by density-based outlier detection, which is usually observed in information-rich areas, and then localizes the location of non-outlier signature point by sparse representation to form the fusion gait template. Identifying users with gait features converted from the gait template showed to be possible. Experiments with a dataset of 175 subjects show that the proposed algorithm significantly outperforms other existing methods and achieves a high recognition rate of 98.67% with single accelerometer of right pelvis.