パーティクルフィルタと単眼視点固定カメラを用いた 3 次元空間上での物体領域推定

Estimating areas of moving objects on 3D space using particle filter and single fixed-viewpoint camera

Creators Yoshimoto Yusuke

Creators Okamura Kenshiro

Creators Matsumura Ryo

In recent years, demands for video surveillance system have increased. Usually, a multi-cameras system is used to have some 3D information in video surveillance system. But such a system needs much cost and time to be put in place. For a low-cost system, we prefer to use only one fixed-view point camera which is calibrated using the projective geometry which projects 3D points onto 2D image. But using only one camera, infinite points on 3D space correspond to a single point on 2D image. In this paper, we propose a
high performance method for estimating areas of moving objects on 3D space using one fixed-viewpoint camera and particle filters. Using some constraints on 3D size and position of objects, our method restricts the generation of particles which represent the candidate area of the object. Therefore, the distribution of the particles shows the appropriate estimation of the object.

[ISSN]0387-9232

[NCID]AN00031668