Simultaneous Localization and Mapping using an Omni-Directional Camera

The ability of a robot to determine its position while at the same time creating a map of a previously unknown environment (also termed SLAM) is an important step towards autonomy in mobile robotics. One well-known approach consists of estimating the robot position and orientation, as well as the map comprised of the positions of landmarks in the environment, using an Extended Kalman Filter (EKF). Commonly found in the literature are implementations using sensors, such as laser scanners, that return range and bearing measurements to the landmarks. Cameras, on the other hand, which are less expensive and more robust, only yield bearing measurements. This imposes additional challenges on the SLAM algorithm, particularly for feature initialization and data association. In this paper, we propose an implementation of bearing-only SLAM in an indoor environment using an omnidirectional camera as primary sensor. We will discuss Computer Vision algorithms for image-unwrapping and line-extraction, as well as EKFbased methods for bearing-only SLAM, followed by results from extensive simulations and initial hardware experiments on a PackBot robot.