Feature detection and map building using ranging sensors

Abstract

This paper describes two methods for feature extraction. The goal is to interpret the data obtained from sonar measurements of the robot's environment as geometric map primitives. The extracted primitives can be used to update the robot position and the global map. This is the process how the robot can explore an unknown environment. The first algorithm gathers sonar measurements into sensor-based map and then using computer-vision techniques and morphological operations extracts geometric description of detected obstacles. The second method is started whenever the border tracking algorithm has been activated. This method uses raw sonar data as the input while obstacle border is robustly detected by the border tracking process. The algorithms have been verified with an experimental mobile robot at the Czech Technical University.