Three-dimensional place modeling and recognition for mobile robots

On robot navigation, location information is necessary for a robot to execute right behavior at correct places. General navigation systems utilize a map of space or a graph-representation from passive learning or man-made data input. But for space representations to be generated automatically by a robot itself, the ability of recognizing the places it has explored become more crucial. The method of place recognition the authors are presenting is a fast and effect way for a robot to recognize place from a single image. The robot moves in the region of a place and models the place by a set of images that it capture at different viewpoints. Correspondences between segment features of the image to be recognize and those of place models are established through some geometric constraints and invariants. Simulation is implemented to support their idea theoretically and the results of experiments show the practical performance of this method