Roland Geibel 
  
SEGMENTATION OF LASER ALTIMETER DATA FOR BUILDING RECONSTRUCTION: 
DIFFERENT PROCEDURES AND COMPARISON 
Roland GEIBEL, Uwe STILLA 
FGAN-FOM Research Institute for Optronics and Pattern Recognition 
D 76275 Ettlingen, Germany 
{gei,usti } @fom.fgan.de 
  
Working Group III/3 
KEY WORDS: Segmentation, Laser, Building, Evaluation, Surface Reconstruction. 
ABSTRACT 
One important step in the reconstruction of buildings from laser altimeter data is the segmentation step. In this 
contribution four procedures known from literature are investigated and compared. One procedure, which was made for 
detecting straight lines, which produces a segmentation as an intermediate result, was used. Three other procedures 
were installed, which are based on region growing. For each of the procedures experiments with an extensive variation 
of the parameters were carried out. For the particular task of roof reconstruction and the speciality of height data an own 
procedure was developed. In this procedure the distance measure is not computed from derived properties but directly 
from the sets of points. Moreover it also uses the knowledge about the separation of background and foreground and it 
is regulated by only one parameter. In order to make the segmentation results comparable known comparison measures 
for single segments (the relative correspondence related to the model and the result) were taken. For the estimation of 
the whole segmentation several aspects of the evaluation of single segments are discussed and a complete evaluation 
function is developed. For a quantitative measurement some buildings were chosen as examples. 
1 MOTIVATION 
1.1 City models 
Airborne height data recorded with laser scanners are gaining increasing importance in generating three dimensional 
city models. There is a particular demand for such models in visualisation, mission planning, disaster management and 
as a base for simulations, e.g. the area of environment protection and telecommunication. In industrialised countries in 
recent years many maps were recorded digitally and they are more and more available as vector maps. However large 
scale topographic maps and cadastral maps show only the footprints of buildings without information about the height 
or the roof shape. So far information about the height of buildings was derived by manual measurement or from stereo 
image pairs. 
At the present time height data of airborne laser scanner systems are commercially available (e.g. [Lohr, 1998] [Huising 
& Gomes Pereira, 1998]). The sensed surface points, scattered over a stripe of 250-500m width, allow the production of 
a geocoded 2D-raster with height data for each cell (height image). Single flight strips are combined into a consistent 
digital height model of the area investigated. 
A main component of city models is the vector description of the buildings with their roofs. For analysing the roof 
shapes the height data belonging to each building are masked using a digital map. For the reconstruction of the roof 
shape the segmentation of the height data as the first processing step plays an important role. For the segmentation of 
range data, several procedures have already been developed. In a former paper procedures for laser scanner images and 
structured light images of some polyhedral bodies were investigated in a laboratory-like environment. It still is an 
unsolved problem, which of these procedures is most appropriate for the task presented here and the specific data of 
natural scenes with artefacts. 
1.2 Related Work 
Using height images produced by laser scanner data, a big part of the roof shapes usually found in urban areas can be 
reconstructed. Thus [Stilla et al., 1998] dealt with the reconstruction of different roof shapes (sloped and flat, with and 
without superstructure). Therein a building with sloped and flat roof parts could be reconstructed using a segmentation 
of the height image and of the gradient image. Thereon it was used that roofs of rectangular buildings usually have 
  
326 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
  
rn 
dde ade W^SCM oum 
es 
B cL 
— A -—— À —  À atl Raub anb Gab 5 da " Fay Hs oi on und — 5S