Visual impression localization of autonomous robots

This paper proposes a novel localization approach based on visual impressions. We define a visual impression as the representation of a HSV color distribution of a place. The representation uses clustering feature (CF) tree to manage the color distribution and we propose to weight each CF entry to indicate its importance. The method compares the navigating tree, which is created by the robot from its observations, with the available reference trees of the environment. In addition, we propose a new similarity measure to compare two CF trees which represent the visual impressions of the corresponding two places. The method is tested on two data sets collected in different environments. The results of the experiments show the effectiveness of the proposed method.