SUB-SURFACE GROWING AND BOUNDARY GENERALIZATION FOR 3D BUILDING RECONSTRUCTION

Abstract. The automatic reconstruction of 3D building models from airborne point cloud data is still an ongoing research topic. Especially for complex roof shapes, the identification of sub-shapes, the generation of roof boundaries and the construction to well-shaped and topologically correct models remains only partially solved. In this paper, a 3D building reconstruction methodology that is based on the notion of sub-surface growing as a means for point cloud segmentation of planar surfaces is introduced. In contrast to conventional surface growing, the segmentation process continues below other surfaces. As a result, the segments grow larger, their number decrease and the adjacency relations between them become more distinct, thus allowing stricter rules to help identify and differentiate between the root types of the roof sub-shapes and their composition to complex roof structures. In conjunction with a constructive solid modeling approach, the model construction is significantly simplified, as the generation of primitives from sub-surface segments is straightforward and their combination to complex shapes can be much easier derived from their interrelations. In the second part of the paper, a boundary generalization approach is presented that allows generating building and segment outlines with regularized shapes from given point sets. Together with sub-surface growing, its usage in the reconstruction of flat roof office buildings is shown. The models are constructed in layers in a bottom-up fashion, each one being the result of a flat sub-surface segment with a generalized boundary, where the regularization rules of one layer are propagated to the next in order to gain well-shaped buildings. A discussion on the so far achieved results and future developments concludes the paper.