Vision for a Robot Moving in an Indoor Environment

Abstract

This paper presents four complementary research contributions used to provide a representation of the 3D indoor environment of a mobile robot. The first tool needed to produce adequate information is the segmentation algorithm: we choose to perform a region segmentation according to a region growing method. The perception of the distance is done by using two cameras: we have implemented a stereovision method founded on region matching to reconstruct 3D surfaces. We have developed a monocular analysis to extract the perspective information from the images. Finally, parts of all this information need to be structured in order to be able to use themThe result of all this work is a rich collection of partly structured 2D and 3D information, which represents the first step towards model learning so that the robot may be able to locate itself to navigate and to identify the objects surrounding itselfThe four processes developed are inter-dependent; for example, the structuration uses the vanishing points extracted by monocular analysis and the 3D segments produced by stereovision depend on the segmentation results