Frame matching for human identification

Gait recognition has become one of the common practices in multi-class human identification. It has led a research stream to explore the dynamics of an entire silhouette sequence. In this paper, a more intuitive human identification approach using direct comparison of silhouette features extracted from a single frame is proposed; we refer to this comparison as frame matching. The obtained recognition vectors from all probe frames are then fused together to yield the final output. The proposed approach is evaluated on the standard CASIA Gait Databases (Dataset B) and yields encouraging results.