Are Intermediate Views Beneficial for Gait Recognition using a View Transformation Model?

Gait recognition is one of behavioral biometrics and has an advantage over the other biometrics in terms that it can be used even at a distance from a camera. Accuracies of the gait recognition, however, degrade if observation views of matching pairs of gaits are different. In order to suppress the accuracy degradation, a family of view transformation models (VTMs) for the gait recognition have been proposed, where a gait feature from one view is transformed to that from another different view so as to match the gait features under the same view. Although the transformation view of the VTM approaches can affect the authentication accuracy in general, its effect has not yet been well investigated in previous work. In this paper, we therefore evaluate the effect of the transformation view for gait recognition experimentally, and report the evaluation results using a publicly available large-population gait database.