カルマンフィルタとファジー推論を用いた単眼視時系列画像からの3次元人体運動の再構成

概要

論文の詳細を見る
A system which reconstructs 3D human movement from a stream of 2D input images is developed. The system finds the 3D shape and motion of the human body by fitting a simple skeleton model to the notable regions found in the 2D image. Surface markers which robust for obstruction were applied. The fitting is done using a nonlinear optimization approach. This optimization is essentially ill-conditioned because depth information can not be obtained from 2D image. In order to overcome this problem, Kalman filter using Fazzy reasoning which takes account of following a priori information is applied: 1 The versatile ekeleton model exhibiting a human body. 2 Postures at each keyframe for the type of human movement. We reconstructed human movement from a stream of Video images by this method in order to demonstrate its applicability.
社団法人日本機械学会の論文
1998-08-25