3D scene representation using a deformable surface

Abstract

This paper presents a method based on a deformable surface (DS) for constructing representations of real 3D scenes from multiple range images. The DS model uses physically based surface and edge constraints to reconstruct a representation from noisy 3D data. The representation constructed identifies consistent surface regions and discontinuities which fit the 3D data to within known tolerances. Fusion of multiple range images enables a complete 3D scene reconstruction. A multi-scale representation is obtained which gives an order of magnitude data reduction. Representations can be used for feature matching or CAD model building