Adapting Geometrical Deformable Models to Multiple Range Images

Recent research topics deal with the problem how to reconstruct real-world objects from a set of multiple range images showing only portions of the object. Here we present an approach how to generate and adapt a geometrical deformable model (GDM) to a set of already registered range images in order to reconstruct the complete object. From the range images we derive a signed distance function which implicitly defines the surface of the object. Then, an intermediate volume is carved out and a sparse triangle mesh is generated. The proposed GDM scheme refines the initial roughly approximated mesh by deformation and adaptive subtriangulation. Due to the adaptive improvement of the mesh up to the desired degree of accuracy, our method describes an efficient way how to reconstruct the object in user-definable accuracy.