把持の動作モデルに基づくロバストな手指のカメラトラッキングの提案(人工現実感及び一般)

概要

論文の詳細を見る
This paper proposed a camera-based finger tracking system for direct manipulation of virtual objects, that reduces influence of occlusion using grasp motion models. Natural Position (NP) Model was defined as the correlation among the finger positions during five-finger-grasp. If the target finger is completely occluded, its position is estimated using NP Model using the positions of neighbor fingers. Potential Energy (PE) Model was also defined to describe the anti-mobility of a finger from its NP. If one camera succeeds to track the finger, the position that requires minimum PE from NP is calculated along the camera optical line. The position estimation error of supposed occluded finger was less than 30mm in grasping fingers except thumb. The error was reduced to 10mm in single-camera tracking case.
社団法人電子情報通信学会の論文
2007-05-28