Real-time structure from motion for monocular and stereo cameras

This paper deals with the real-time estimations of the rigid motion and three-dimensional structure of a scene from one or stereo moving camera. The method is based on an Extended Kalman Filter (EKF) that permits to obtain an executable real-time implementation. The extensions of an existent single camera solution and the new formulation of the stereo case is explained. Synthetic and real data experiments show that the estimations obtained by the binocular system are more accurate and the system is more robust as compared to the monocular case.