Detecting high level features for mobile robot localization

Robust mobile robot localization requires the availability of highly reliable features obtained by the external sensors of the robot. Redundancy assures reliability and precision of the observed features. In this work we use two different sensors, namely, a laser rangefinder and a monocular vision system, whose complementary nature allows one to robustly identify high level features, i.e. corners and semiplanes, in the environment of the robot. We present a general fusion mechanism, based on the extended information filter, supported by a robust modelling of uncertain geometric information, to fuse information obtained by different sensors mounted on the robot. Localization of the robot is achieved by matching these observations with an a priori map of the environment. An a priori estimation of the robot location is not required. Experimental results are presented, showing the increase in reliability of the observed features after fusing information from both sensors