画像に基づく歩容解析 : 個人認証及び属性推定

Abstract

Image-based gait analysis methods proposed by authors' group are described in this paper. Silhouette-based gait features are extracted from gait image sequences, and the extracted features are used for person verification tasks and person attribute estimation tasks such as age and gender. As the gait verification tasks, approaches for issues of view difference, belonging difference, and occlusion are explained. For the view issues, view transformation-based approaches and CNN-based view-invariant approaches are explained. For the issue of belonging difference, joint intensity and spatial metric learning method is briefly explained. And for occlusion issue, subspace-based gait regeneration methods and a WGAN-based silhouette image sequence reconstruction method are explained. And as an application for gait verification, gait verification system for criminal investigation is introduced.