A Split-and-Merge Technique for Automated Reconstruction of Roof Planes

Abstract

Automated reconstruction of buildings from different data sources has been one of the most challenging problems in photogrammetry and computer vision. Systems for automated building reconstruction fail in many cases due to complexities involved in the data including image noise, occlusion, shadow, and low contrast, as well as, low accuracy or density of height data. In this paper, the problem of overgrown and undergrown regions in the segmentation of aerial images is discussed, and a split-and-merge technique is presented to overcome this problem by making use of height data. This technique is based on splitting image regions whose associated height points do not fall in a single plane, and merging coplanar neighboring regions. A robust plane-fitting method is used to fit planar surfaces to height points that are highly contaminated by gross errors. Final roof planes are extracted out of the image planar regions by checking their slope and height over a morphologically opened DSM. An experimental evaluation is conducted, and its results indicate the capability of the proposed technique in splitting overgrown regions, merging undergrown coplanar regions, and selecting the final roof planes. Also, the method is shown to be computationally efficient, and the reconstructed roof planes are of acceptable accuracy.