Hiroshi Masaharu 
  
THREE-DIMENSIONAL CITY MODELING FROM LASER SCANNER DATA BY EXTRACTING 
BUILDING POLYGONS USING REGION SEGMENTATION METHOD 
Hiroshi MASAHARU and Hiroyuki HASEGAWA 
Geographical Survey Institute, Japan 
Geographic Information Analysis Research Division 
masaharu @ gsi-mc.go.jp 
hase @gsi-mc.go.jp 
Working Group V/3 
KEY WORDS: Laser/Lidar, Urban objects, DTM/DEM/DSM, Modeling, Buildings, GIS, Data acquisition 
ABSTRACT 
We have developed a method to automatically generate a 3-D city model only from laser scanner data by applying 
region segmentation to the grid type data of digital surface model (DSM) obtained by the scanner. Each building has a 
distinct height difference from the surroundings. Therefore we can distinguish each buildings by segmenting the DSM 
with the condition that a pixel having the height difference within a predetermined threshold from a neighboring pixel 
belongs to the same region. 
Results are given for parts of two different type cities, Minokamo and Tsukuba, Japan. We evaluate from these results 
that this method is useful to promptly obtain 3-D city models in vector format because the model is generated through 
automatic processing of the data. But the resulting model is not yet suited for direct usage in GIS because the resulting 
polyhedrons obtained by this method correspond to parts of buildings and not necessarily to each building. It is suited 
in particular for 3-D scenery reconstruction as can be seen in the figures of the bird's-eye view. 
This method, however, has its intrinsic problems that in some condition even large buildings cannot be distinguished 
from road surface. This means that these buildings disappear in the obtained 3-D model. Another problem is that the 
extracted objects are not always buildings or houses but, for example, trees are included. These issues are also 
discussed in the paper. 
1 INTRODUCTION 
Three dimensional city models handled in a 3-D GIS environment are attracting much more attention due to the 
potential of its applications to many fields and increasing availability of 3-D GIS environment. For that purpose, it is 
necessary to create 3-D city model data. Airborne laser scanner is one of the most promising tools that measure three- 
dimensional shapes of ground objects. But it provides in principle only the x, y and z coordinates of points densely 
distributed on the surface of ground objects, or z values at dense regular grid interval (grid DEM type data). Therefore 
it is necessary to reconstruct appropriate geometric models representing buildings, houses and other ground features 
from these data. 
A number of researches have been carried out to make detailed 3-D building models by using both 2-D digital map data 
and laser scanner data (Haala and Brenner, 1999; Maas and Vosselman, 1999; Stilla and Jurkiewicz, 1999). As 2-D 
digital map data in vector format are available in many cities nowadays, this method would be useful and applicable. 
But changes taking place between digital map data provision and laser scanner data acquisition are problem in this 
method. 
We started to develop methods in which only the laser scanner data are used for reconstructing 3-D city model so that 
we can explore the possibility of this sensor in obtaining three-dimensional spatial data. One of the merits of the laser 
scanner sensor is that it can provide digital data within a few hours after the flight. This swiftness will be useful for 
many applications. In this regard, we consider this approach is advantageous because no other data than laser scanner 
data are necessary and the processing will be done automatically in principle. 
We have developed a 3-D city model generation method based on region segmentation of laser scanner data. The 
results are presented and the problems of the method and future study items are discussed. 
  
556 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
The n 
by ap 
height 
surfac 
neighl 
Bounc 
by thi 
given 
2.1 | 
The r« 
agains 
same 
used | 
regule 
Befor 
origin 
2.2 ] 
Label 
segme 
of ve 
algori 
are ex 
2.3 
Next 
apply 
scann 
3.1 
The « 
frequ 
Cross 
width 
speci 
CCD 
simul 
featui 
featu 
3.2 
Mino 
centr 
numt