Which Reference View is Effective for Gait Identification Using a View Transformation Model?

Gait identification is a promising method of individual identification at a distance from a camera and identification of those who observed from various views or those who going to various directions is required in particular for actual use. In this paper, we discuss a selection of reference views for the various-view gait identification using a view transformation model (VTM). In the gait identification process, we first extract frequency-domain gait features from gait silhouette sequences, and then obtain the various-view gait features by transforming a few reference features with the VTM. We made experiments using 736 sequences from 20 subjects of 24 view directions. We evaluate the performance for each single reference and for each combination of two references. In addition, we inspect the relation between the performance and the number of references.