Istanbul 2004 
vanced DTM 
XIV — 3/W4, 
h, DIGITAL 
R SCANNER 
ute of Digital 
ustria. 
ger, (2002). 
.ATION FOR 
rammetry and 
Technology, 
zy Summary: 
html. 
98. 3D urban 
PRS. Vol. 32, 
Henry, Wide- 
sing Airborne 
logy PE&RS 
Iter to Extract 
esy , Stuttgart 
  
  
    
  
3D BUILDING MODEL GENERATION FROM AIRBORNE LASERSCANNER DATA BY 
STRAIGHT LINE DETECTION IN SPECIFIC ORTHOGONAL PROJECTIONS 
Ellen Schwalbe 
Institute of Photogrammetry and Remote Sensing 
Dresden University of Technology 
Helmholtzstr.10 
D-01062 Dresden, Germany 
ellen.schwalbe@mailbox.tu-dresden.de 
Commission III / WG3 
KEY WORDS: LIDAR, point cloud, building model reconstruction, automation, visualization 
ABSTRACT: 
The paper describes a novel approach for the generation of 3D building models from airborne laserscanner data, which is based on 
the detection of straight lines in specific 2D projections of the data. First, the main roof ridge directions of a building are 
determined. Based on these ridge directions, hypotheses on the roof orientation are generated. The whole point cloud is rotated by 
the roof orientation and orthogonal projected into a 2D space perpendicular to the roof orientation. Roof edges show up as straight 
linear point clusters in this 2D projection. These lines representing projections of roof faces are extracted by a line tracing 
technique. They provide information about inclination and width of the roof faces. Subsequently rotating the points of each roof 
face around the z-axis and tilting it by its inclination produces another 2D projection containing lines representing the roof face's 
length. The technique was tested on a sample of 200 buildings. They were reconstructed with the approach and evaluated for their 
correctness and geometric accuracy. 
1 INTRODUCTION 
1.1 Motivation 
The topic of building model reconstruction from airborne laser 
scanner data has become more interesting in the past years. 
Examples of the various application fields are the generation of 
3D-city models for city planning or virtual city tours as well as 
the acquisition of information for geographical information 
systems (GIS) and their derived products. Furthermore, 
building modelling can be used for the revision of cadastral 
maps. 
For this reason one searches for methods with which buildings 
can be modelled automatically from laserscanner data (see 
Figure 1-1) as reliably and effectively as possible. Considering 
the increasing point densities and accuracy of the laserscanner 
data, it is also of interest to develop methods which are scalable 
in a way that they produce a level of detail that depends on the 
density and quality of the data. 
Under these criteria different approaches had been developed 
for building model reconstruction. Methods, which are based 
purely on laserscanner data, can be differentiated into model- 
driven approaches such as [Maas 1999] and numerous data- 
driven approaches. The data-driven approaches have been 
realised by using for instance a segmentation procedure e.g. 
[Rottensteiner 2002] or a region-growing-based method e.g. 
[Gorte 2002]. Another data-driven approach, which is 
parameter-based, was introduced by [Vosselmann 1999]. 
  
  
Figure 1-1: Comparison of the derived building model with the 
original building