A Head-Wearable Short-Baseline Stereo System for the Simultaneous Estimation of Structure and Motion

H. Badino and T. Kanade


This paper presents a short-baseline real-time stereo vision system that is capable of the simultaneous and robust estimation of the ego-motion and of the 3D structure and the independent motion of thousands of points of the environment. Kalman filters estimate the position and velocity of world points in 3D Euclidean space. The six degrees of freedom of the ego-motion are obtained by minimizing the projection error of the current and previous clouds of static points. Experimental results with real data in indoor and outdoor environments demonstrate the robustness, accuracy and efficiency of our approach. Since the baseline is as short as 13cm, the device is head-mountable, and can be used by a visually impaired person. Our proposed system can be used to augment the perception of the user in complex dynamic environments.