Automatic Reconstruction of Roofs from Maps and Elevation Data

In this paper we describe a procedure for generating building models from large scale vector maps and laser altimeter data. In contrast to other approaches the map information is not used to form hypotheses about the roofs. Our approach permits the recognition of additional structures as superstructures which can not be derived from building outlines shown in the map. First the vector map is analyzed to group the outlines of buildings and to obtain a hierarchical description of buildings or building complexes. The base area is used to mask the elevation data of single buildings and to derive a coarse 3D-description by prismatic models. Afterwards, details of the roof are analyzed. Based on the histogram of heights, flat roofs and sloped roofs are discriminated. For reconstructing flat roofs with superstructures, peaks are searched in the histogram and used to segment the height data. Compact segments are examined for a regular shape and approximated by additional prismatic objects. For reconstructing sloped roofs, the gradient field of the elevation data is calculated and a histogram of orientations is determined. Major orientations in the histogram are detected and used to segment the elevation image. For each segment containing homogeneous orientations and slopes, a spatial plane is fitted and a 3D-contour is constructed. In order to obtain a polygonal description, adjacent planes are intersected and common vertices are calculated.