ARToolkit Applied to Panoramic Vision for Robot Navigation

A vision based navigation system is presented for determining a mobile robot’s position and orientation using panoramic imagery. An omni-directional image sensor mounted on the robot is useful in obtaining a Æ field of view, permitting navigational markers from all sides to be viewed simultaneously. In this novel approach, navigational markers are recognized using software from the popular ARToolkit software package. These marker patterns are arrayed in the environment and recognized in quasi-perspective projection views warped from the panoramic image. A novel way is shown for generating these perspective projection images from the non-perspective image seen by the panoramic camera, without relying on the Single Viewpoint Criteria which greatly restricts the camera design. An a priori model is used to predict the marker positions, from which quasiperspective views are generated. The ARToolkit software operates on these views and identifies and locates the markers’ directions from which the robot’s position is updated. Robust performance is achieved by the uniqueness of the markers as landmark features. Experiments are shown with synthetic imagery.