A feature-level approach to multi-sensor integration; emphasizing robotic applications

Abstract

Employing a feature level description of the robotic environment the two important issues of what to fuse and how to fuse are developed in detail. The former is known as the registration (or correspondence) problem, and the latter the fusion problem. The registration procedure is addressed by a comprehensive two stage process, The first stage affords the development of local sensor interpretations which are then registered between sensors and fused to construct a global interpretation. To perform the actual fusion, information-scaled estimates are employed to develop a decentralized fusion structure; temporal and spatial alignment is also supported.<>