プログラマブルシェーダを活用した3次元ビデオアバタの生成(人工現実感)

概要

論文の詳細を見る
To communicate with remote user from different area, we proposed 3D video-based avatar, which provide the remote user's human figure from various viewpoint in realtime. One popular method for 3D geometrical model recognition from multi-camera images is volume intersection method. However, boxel-based volume intersection algorithm exhausts a massive amount of calculation power and memory space. So, it is difficult to generate high-resolution result image in realtime. In this paper, we propose a frame-buffer based intersection algorithm based on goldfeather method to evaluate intersection volume of 3D objects in realtime. By using programmable shader, it is easy to execute user's programmed frame-buffer processing and rendering algorithm for each frame on graphic hardware. By applying the proposed algorithm on GPU, we can generate and render 3D video avatar from multi camera images in realtime.
2005-05-27