A C-embedded algorithm for real-time monocular SLAM

This paper describes a new architecture for Visual Simultaneous Localization And Mapping (SLAM), aimed at being implemented on different embedded boards on mobile robots. We detail the architecture of our C-coded program, which gives real-time results. We present the different methods used to get a real-time solution for the visual SLAM problem, for image processing, landmark parameterization and reparameterization, Extended Kalman Filter (EKF) and Active Search. We finally expose the results of our algorithm embedded on three different boards which are intended to be mounted on mobile robots.