Omnidirectional vision for robot localization in urban environments

This paper addresses the problem of long term mobile robot localization in large urban environments. Typically, GPS is the preferred sensor for outdoor operation. However, using GPS-only localization methods leads to significant performance degradation in urban areas where tall nearby structures obstruct the view of the satellites. In our work, we use omnidirectional vision based techniques to supplement GPS and odometry and provide accurate localization. We also present some novel Monte Carlo Localization optimizations and we introduce the concept of on line knowledge acquisition and integration presenting a framework able to perform long term robot localization in real environments. The vision system identifies prominent features in the scene and matches them with a database of geo-referenced features already known or integrated during the localization process. Results of robot localization in the old town of Fermo are presented. Results show good performance and the whole architecture behaves well also in long term experiments, showing a suitable and good system for real life robot applications.