Recognition and localization of generic objects for indoor navigation using functionality

Abstract

Autonomous indoor navigation requires a robot to have abilities to recognize landmarks and to avoid obstacles. Our goal is to provide these abilities in a generic way so that the robot need not have an accurate and complete geometric model of all the objects in its environment. Rather, we wish to provide abilities to recognize common objects such as desks and doors which can be used as landmarks. We use a functionality-based representation for objects. Objects consist of functional parts, which are characterized by their significant surfaces, and by the accompanying functional evidence for them. Our system works with planar surfaces only and assumes that the objects are in a ‘standard’ pose. The localization and orientation of an object are represented with the most significant surface in an ‘s-map’. Our system has been tested on a number of real scenes where it performs robustly and efficiently; some results are shown in the paper.