2A1-A24 モデルベースドモーションキャプチャのための個体適応人体モデルの構築

概要

論文の詳細を見る
This paper introduces a person-adaptive model for markerless motion capture system. Various model-based motion capture systems, which estimate human motion by comparing the 3D human model with either a reconstructed 3D human shape or captured images, have been proposed so far. However, a unique 3D human model to each person has to be prepared in advance, since joint positions and size of the 3D human model cannot be changed during the processing of estimating one's motion. To overcome this problem, 3D human models using cylinders or superquadric ellipsoids have been proposed, but there are still problems to be solved, for example, human motion cannot be estimated precisely due to the shape difference between the actual human shape and the 3D human model, and the number of parameters is large. So in this paper, we propose a parametric 3D human model to estimate human motion precisely. In experiments, we show various 3D human models can be expressed with the change of parameters of the proposed model.