Cross-View Gait Recognition in 3-D Space Based on Gravity Center Track

Abstract

We propose in this paper a novel cross-view gait recognition method based on gravity center trajectory (GCT). Inspired by the finding that if the GCT of human in walking process has regularity, the representation coefficients of the trajectory are generally consistent across different views. We propose to project the coefficients of GCT to different view plane (VP) which is the normal plane of view angle direction vector to achieve view-invariant features for gait recognition. Firstly, we obtain the GCT under different views by summation of pixel coordinates in body area. Then, we use the least square method to eliminate the upward or downward trend of GCT caused by view variance. Then, we project the GCT function to the corresponding VP. Lastly, we perform recognition by using a simple cluster method. Experimental results on the widely used CASIA-B gait database demonstrate the effectiveness and practicability of the proposed method.