The Caboto Project

This dissertation describes the implementation of an omni-directional vision system as sole sensor for a robot building a topological map of an indoor environment, using the Spatial Semantic Hierarchy created by Benjamin Kuipers [Kuipers 00]. So far, no attempt to implement the SSH using a vision sensor has been made. We will show that an omni-directional vision system is a good sensor for the SSH. We built a simulator for the omni-directional vision system. Using this simulator, we created typical omni-directional images of an indoor man-made environment. From a qualitative analysis of these images, we identified the transitions in the image sequence and the features in the images them-selves which are strictly related to the SSH representation. We designed algorithms to extract this information from the images. Finally, we tested these algorithms with simulated experiments and using a real robot fitted with an omni-directional vision sensor. The omni-directional sensor is realized with a multi-part omni-directional mirror and a colour camera under the mirror. We used the colours to solve the correspondence problem in the frame sequence. Our approach worked properly in simulation but was not robust enough to cope with very noisy images in the real world. More sophisticated solutions are outlined. We present detailed descriptions of the image processing and of the matching algorithm used for the solution of the correspondence problem.