幾何学的位置合わせのための自然特徴点ランドマークデータベースを用いたカメラ位置・姿勢推定(「複合現実感3」特集)

概要

論文の詳細を見る
In the field of wearable augmented reality, absolute position and posture of user's camera are necessary to overlay CGs on real images. In this research field, two kinds of approaches have been investigated; sensor-based and vision-based methods. However, both kinds of conventional methods have a problem that it was not easy to use these methods in large and variable environments; usable environment is limited by selection of sensors in the sensor based method, and many artificial markers must be allocated in large environment in the vision based method to prevent accumulative errors. This paper proposes a novel vision-based camera position and posture estimation method which does not require sensors and artificial markers. In the proposed method, feature landmark database is constructed in advance by using large number of natural feature points in omni-directional image sequences. Next, by detecting feature landmarks from images captured by wearable camera, camera position and posture are estimated in every frame without accumulative errors.
日本バーチャルリアリティ学会の論文
2005-09-30