Using multiple view geometry within extended Kalman filter framework for simultaneous localization and map-building

One of the recent and consistently interesting topics in robotics research community is the simultaneous localization and map-building (SLAM) problem. It examines the ability of an autonomous mobile vehicle starting in an unknown environment to incrementally build an environment map and simultaneously localize its pose within this map. In this paper, we present a solution to the SLAM problem with minimal initial knowledge. The novelty lies in its monocular vision sensing system, which uses a multiple view geometry (MVG) approach within an extended Kalman filter (EKF) framework. The MVG algorithm provides accurate structure and motion measurements from a monocular camera whereas traditional vision-based approaches require stereo-vision. It is evident from simulation results that the limitations of MVG and EKF, when used on their own are overcome in the proposed solution.