Vision-based metric topological SLAM

In this paper, we present a vision-based metric/topological approach for the Simultaneous Localization and Mapping (SLAM) problem. Similar to recent works on metric/topological SLAM (MT SLAM) where the map is represented as a graph whose nodes are submaps and edges are relative transformations between submaps, we propose a graphical model for vision-based MT SLAM that estimates the hybrid robot pose and the map by using Bayesian filtering. We also propose a motion estimation method that incorporates the uncertainties of 3D landmarks to improve the accuracy of motion estimation and apply this method to our SLAM approach. We demonstrate the performance of the proposed approach using a stereo camera.