Surface-Based Structure-from-Motion using Feature Groupings

Abstract

In this paper we describe a complete system from feature extraction to reconstruction of 3D models of indoor environments. The system uses a novel matching algorithm which matches groupings of features associated with boundaries of objects in the scene. We also present an extension of our structure-from-motion algorithm to incorporate surface constraints. We describe how planar surfaces can be incorporated into the model update procedure, and are hypothesised from the matched groupings of features between image frames. We present reconstructions of environments taken by an autonomous robot to demonstrate the improvement that can be achieved by this approach. Keywords: Environment modelling, structure from motion, feature matching, mobile robotics 1. INTRODUCTION In this paper we consider the problem of reconstructing 3D indoor environments, in near real time, using an autonomous mobile robot. The reconstruction is based upon a structurefrom -motion algorithm using data from feature ...