Abstract
Humans build a cognitive map by extracting geometry and semantic information from the real world. This map plays a significant role in navigation and communication. In this paper, we propose an algorithm for mobile robots for building cognitive maps autonomously. The robot observes the environment with a stereo camera to detect regions that are likely to include text information. If the robot finds such region, the zoom camera zooms in that region to estimate the probability of text existence more precisely. If the probability is high, the robot approaches there and extracts the text information by using the proposed algorithm. The robot builds a cognitive map by storing both text and geometry information. We conducted several experiments to show the validity of our proposed algorithm.
