Cross view gait recognition by metric learning

In this paper, we propose the human recognition framework based on the biometric trait conveyed by a walking subject, where the viewing angles of gallery and probe may differ. To deal with this kind of intra-class variance, we propose to exploit the view transformation technology to transform the embedded vector of one viewing angle into another embedded vector of target viewing angle. Then, the metric already learned previously on target manifold will be use to measure the similarity between vectors. Experiments were conducted on CASIA-B gait database and the results demonstrate the notable improvement of cross view gait recognition performance via the combination of feature transformation and metric learning.