Topological Localization for Mobile Robots Using Omni-directional Vision and Local Features

This paper addresses the problem of global localization (“where am I”?) by a mobile robot using an omni-directional vision sensor. The approach is mainly concerned with the problem of global localization in a topological map (consisting of a set of panoramic views), though it also gives an approximate estimate of the robot pose relative to the best matching view, so it could also be used as the basis for a geometric localization scheme. Four different methods for selecting features from the panoramic images were considered. Experiments were conducted in a large, populated indoor environment on different days under different lighting conditions. A separate, laser-based SLAM algorithm was used to generate the ground truth information for performance evaluation. The results show that one method based on local features (“modified SIFT” features) obtained a significantly higher level of performance and robustness to environmental variations compared to the other methods.