3-D scene reconstruction from image sequences

In this paper we describe a technique for reconstruction of a static 3-D scene from a monocular image sequence. The problem is formulated as a stochastic filtering problem where state variables describe the scene and the camera parameters, and images are observations of this state. This rather general problem formulation allows us to drop requirements otherwise requested, like knowledge of the camera motion path, or the necessity to have a binocular or a trinocular image sequence. The scene is represented as a set of 3-D line segments, edges contours etc. The computation time of each state update is linear with the number of scene line segments. In the paper the achieved results are given and conclusions are derived.