Thomas Vögtle
Figure 6: Separated 3D objects as building hypotheses, each containing an individual segment number
Afterwards a vectorization is done by means of triangulation networks. Triangles with small perimeters are selected and
merged (Schilling & Vögtle, 1996; Vögtle & Schilling, 1997) which provides 2D contour lines of each segment (Figure
7). The advantage of this method is the closing of interior gaps without affecting (smoothing) the outer contour lines.
After reduction of irrelevant points (Figure 7), e.g. by Douglas-Peuker method (Douglas & Peucker, 1973), these
contour lines can be analyzed regarding shape and size parameters (e.g. rectangular lines, parallel lines, number of
points per area, area etc.). These parameters are used to reduce again the number of non-building objects (e.g. remaining
artifacts or vehicles) and to obtain a better preselection of objects as building hypotheses.
Figure 7: Vectorized contour lines of the objects regarded as building hypotheses
(original and Douglas-Peuker filtered); test building 1 (left) and 2 (right)
4 MODELLING OF BUILDINGS
After preprocessing the objects regarded as building hypotheses are modelled geometrically in three dimensions. Today
several approaches for modelling of buildings exist. An overview of the current status in this field can be found e.g. in
(Forstner, 1999). According to this publication, one of the most important aspects in every strategy is the adopted
building model, i.e. the representation form of building geometry. There are mainly three different groups of building
models: parametric, generic and CSG models.
Using the first representation form, one has to choose a predefined type from a limited set of parametrically described
basic house types (e.g. gable or hip roofed ones). The parameters of this basic forms are adjusted to the regarded
building. In the generic attempt there is no limitation to special geometries of man-made objects. Here a building is
described more generally by a set of connected planes. Therefore, each geometric object consisting of planar surfaces
can be reconstructed. CSG method combines so-called building primitives to complex house models, e.g. parametrically
described cuboids or prisms. The presented approach belongs to the second group, polyhedral models are used to
describe buildings. This implies to estimate roof planes of houses in laser scanner data, combining them by intersection
to determine contour lines and corner points of the building. Afterwards the resulting lines and points built up a wire
frame model of this object.
930 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part B3. Amsterdam 2000.