Michel Morgan 
AUTOMATIC BUILDING EXTRACTION FROM AIRBORNE LASER SCANNING DATA 
Michel MORGAN, Klaus TEMPFLI” 
“Ohio State University, USA 
Civil and Environmental Engineering and Geodetic Science Department 
Morgan.465 @ osu.edu 
‘International Institute for Aerospace Survey and Earth Sciences (ITC), The Netherlands 
Tempfli @itc.nl 
  
TECHNICAL COMMISSION III 
KEY WORDS: Laser scanning, building extraction, morphological filter, connected component labeling 
ABSTRACT 
Laser scanning is a new technology for obtaining Digital Surface Models (DSM) of the earth surface. It is fast method 
for sampling the earth surface with a high density and high point accuracy. In this paper a procedure for building 
detection and roof extraction from the DSM is presented. The procedure starts by re-sampling elevation as obtained by 
laser scanning into regular grid. The core part of the building detection is based on a morphological filter for 
distinguishing between terrain and non-terrain segments. The non-terrain segments are classified into building or 
vegetation. Aiming at a vector representation of buildings the roof faces are obtained by further segmentation of the 
building segments into sub-segments. The 3D geometrical properties of each face are obtained based on plane fitting 
using leastsquares adjustment. The reconstruction part of the procedure is based on the adjacency among the roof faces, 
Primitive extraction and face intersections are used for roof reconstruction. Prior knowledge about the buildings and 
terrain is needed for both the detection and extraction processes. The procedure is developed to work for all terrain 
types and for many building/roof types. The laser data used in this research have an average density of 2-3 points per 
square meter and 0.10m standard deviation of the elevation values. The procedure shows promising results for building 
detection. The reconstruction part shows promising results for some roof faces, obtaining high planimetric and heigh 
accuracy. Some adaptations of the procedure are recommended for enhancing the performance of the presented 
approach. 
1 INTRODUCTION 
Many applications, such as urban planning, telecommunication, security services ask for 3D city models. Buildings art 
the objects of highest interest in 3D city modeling. Urban areas are usually rapidly changing mainly due to human 
activities in construction, destruction or extension of topographic elements such as buildings and roads. This leads © 
requesting a fast data acquisition technique and automatic method for detecting and extracting 3D topographic objects 
from the data. Airborne laser scanning is a new technology in which several sensors are integrated to obtain 3D 
coordinates of points on the earth. It makes use of precise GPS instruments to determine the position of the sensor 
inertial navigation system (INS) to determine the attitude of the sensor, and narrow laser beams to determine the rang 
between the sensor and the target points. Laser scanning systems are active, therefore they can work day and night 
Shadows also do not affect laser data. 
In this paper an automatic procedure is presented for building detection and roof extraction from the DSM as obtained 
from laser scanning. Complete reconstruction of buildings with vertical walls from reconstructed roofs can be done 
following the approach described in (Tempfli, 1998). The laser data used in this research are described in section 2. The 
procedure for building detection and extraction is described in section 3. In section 4, empirical tests and the analysis 0 
the results is presented. Section 5 includes the conclusions and the recommendations and the possible enhancements of 
the procedure. 
2 DATA DESCRIPTION 
The laser data used in this research are of an area located in Haren in The Netherlands obtained by TopoSys las 
scanner. The data were kindly provided by the Surveying Department of the Ministry of Transportation, Public Works 
  
616 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
and V 
meter 
truth « 
points 
the D. 
the bt 
reliab 
3s 
31 I 
Proce: 
distrib 
two-di 
sensib 
3.1.1 
Unless 
regula 
resulti 
recipi 
densit 
the reg 
values 
raster, 
3.1.2 
Interpx 
chance 
difficu 
buildir 
nearest 
compo 
neighb 
buildin 
3.2.1.1 
Separa 
Wehr, 
each w 
(band y 
weight 
pixels, 
this eq 
along tl 
used to 
the slog