Experiments in Visual-Based Navigation with an Omnidirectional Camera

摘要

This paper overviews experiments in autonomous visual-based navigation undertaken at the Instituto de Sistemas e Rob\\'otica. By considering the precise nature of the robot's task, we specify a navigation method which fulfills the environmental and localization requirements best suited to achieving this task. On-going research into task specification using 3D models, along with improvements to our topological navigation method are presented. We show how to build 3D models from images obtained with an omnidirectional camera equipped with a spherical mirror, despite the fact that it does not have a single projection centre.