Thomas Vögtle
processing. Up to now, aerial colour infrared (CIR) images are used for this but newest developments aim at using
additional multispectral scanner. In future, the integration of spectral information by registering the intensity values of
the laser impulses themselves is planned for laser scanner systems. When these systems will be operational, the use of
additional data sources will not be necessary.
The strategy of this approach is a stepwise elimination of areas and objects which are not buildings. This can be
achieved by spectral, shape and size parameters. In a first step, both the CIR images as well as the elevation data
derived from laser scanning have to be preprocessed. The CIR images are geometrically rectified (orthophotos) by
means of laser DEM. The subsequent identification of vegetation areas is based on well-known vegetation index NDVI
(section 3.1). On the other hand, terrain undulations have to be eliminated from the elevation data, so only 3D objects
on this surface remain (section 3.2). In the next step, all vegetation objects are eliminated by superimposition of NDVI
image with preprocessed elevation data. The remaining 3D objects are separated from each other and their shape and
size analyzed in order to exclude other non-building objects, e.g. vehicles or remaining artifacts (section 3.2). All 3D
objects which are left after this preprocessing are regarded as building hypotheses and therefore modelled by planes
. which are estimated by a special region growing algorithm (section 4.1). The intersection of these planes leads to the
resulting contour lines and corner points of each object. These graphical primitives are used to build up a wireframe or
surface model (section 4.2).
3 PREPROCESSING
3.1 CIR images
For a subsequent superimposition of CIR images with laser DEM both data have to be related to the same georeference
system. While laser scanning data are already geocoded, CIR images have to be rectified to obtain geometrically
comparable data sets. Therefore, CIR orthophotos are produced by means of the laser DEM (Figure 1).
Figure 1: Georeferenced CIR image (orthophoto), Figure 2: Laser dDEM, terrain reduced
subset of test area Karlsruhe (original DEM - DTM)
In the next step vegetation areas are extracted by well-known Normalized Difference Vegetation Index NDVI (Schilling
& Vogtle, 1996; Vogtle & Schilling, 1997), although other classification procedures could be used as well.
_IR—R
NDVI = RAR (1)
IR - reflectance value in the near infrared
R — reflectance value in visible red
928 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.