Qualitative Localization by Full-View Spherical Image

The content of a full-view spherical image is independent of the camera rotation. In this paper we propose a method of qualitative localization of a robot by full-view spherical images. First, a robot is guided by humans and memorizes the rout scenes by capturing a spherical image sequence from a spherical image sensor; then the robot localizes itself by matching the current spherical image with the stored ones in terms of the similarity of the feature vectors computed from the histograms of spherical images. The advantage of the proposed full-view image approach over the lower hemispherical images from conventional catadioptric omnidirectional camera is shown by real-world experiments of both indoor and outdoor environment.