Gait-based human recognition by multiple wavelet coherence

Abstract

In this paper we use Multiple Wavelet Coherence (MWC) for the recognition of human using gait. MWC is analogous to multiple correlation which results coherence of multiple independent signals on dependent one. It describes the region of proportionate wavelet power of independent signals. We extract 1D dependent signals generated due to shoulders and hand movement and independent signals generated due to leg movement. We compute MWC of each sequence of all 20 subjects of CASIA-A gait database walking at an angle 0° to the image plane. Experimental results show that MWC preserves significant discriminant information of walking individual. Finally PCA is used to train the proposed system and for testing we use nearest neighbor method. Cumulative match score is used to evaluate the proposed system.