Omni-directional Visual Navigation

This paper considers a method for the visual-based navigation of a mobile robot in an indoor environment. The sole sensor used is an omni-directional camera. Navigation is achieved by combining appearance based methods and visual servoing. A database of omni-directional images is obtained during a training stage. This database maps the environment in a topological sense. Since the image set is relatively large, it is compressed using Principal Component Analysis. A low dimensional eigenspace is constructed, from which the robot can determine its global topological position along a path. Visual servoing is applied to ground plane dewarped images in order to control locally the pose of the robot. The method of obtaining these images is outlined, as is the closed loop control scheme used to keep the robot moving in the centre of a corridor. Finally, real world experimental results for both static and relatively dynamic environments are presented and proposals made for the extension of the system.