I. Introduction
The degree of autonomy of a robot is very dependent on its ability to perform simultaneous localization and mapping (SLAM). An autonomous robot should be able to explore its environment without user intervention, build a reliable map, and be able to localize itself within that map. Many approaches to the SLAM problem can be found in the literature, but the most promising one in terms of cost, computation, and accuracy is the so-called visual simultaneous localization and mapping (vSLAMTM) [8], [9]. In vSLAM, localization is achieved based on recognizing a visual landmark that is nothing but a set of unique and identifiable feature points in 3D space. If a subset of these features are later observed, the vSLAM system can calculate the new robot's new pose [5].