Entire Shape Recovery of an Object by Virtual See-through Cameras and Its Application to 3-D Modeling of Human Motions

摘要

本論文は,物体を包囲する仮想透視カメラ群を用いて,物体の全周囲形状のモデル化を行う新しい方法について述べている.提案法のカメラシステムは,前面のカメラ群が,物体の前面だけでなく後面も仮想的に観察し, 後面のカメラ群は,物体の後面だけでなく前面も仮想的に観察するというシステムである.これにより,物体全体に広がる特徴点は単一の計測行列に記入され, この計測行列が因子分解されて,全特徴点の3次元座標が求められる.提案法は,リハビリテーションの分野でしばしば解析対象となる,椅子からの人の立ち上がり動作の立体モデル化に適用され,満足のいく結果が得られた.This article describes a novel technique for modeling an entire shape of an object by employing virtual see-through cameras surrounding an object interested. The proposed camera system works as a frontal camera system which observes not only frontal part of the object concerned but also its rear part virtually and, at the same time, as a rear camera system which observes not only rear part of the object but also its frontal part virtually. The feature points spread over the entire object are then collected and registered in a single measurement matrix which is factorized to obtain the 3-D locations of the feature points. The technique is applied successfully to 3-D modeling of a human stand-up motion from a chair which is often analyzed in rehabilitation.