A real-time analysis of 3D scene from monocular images by observing known background

In this paper, we propose a real-time method for analyzing 3-dimensional (3D) scene of objects from time series of monocular images. A single camera observes some points on the object so that 3D position of the points can be estimated by extended Kalman filter. We apply this method to two real-time applications. One is to acquire 3D geometry of an object, and another is to estimate 3D pose/position of an object. This approach needs no model data of the object a priori and achieves the estimation of 3D geometry and pose/position. In the experiments using a manipulation robot, we show that it is effective method to estimate 3D information with high accuracy.