In: Stilla U, Rottensteiner F, Paparoditis N (Eds) CMRT09. IAPRS, Voi. XXXVIII, Part 3/W4 — Paris, France, 3-4 September, 2009 
2.4 Roof Junctions: 
Cells that have neighbor cells at two consecutive sides or at 
three or more sides are examined again. These cells are 
candidates for connecting shapes. Based on the shape types, the 
parameters and the arrangement of the neighbor cells, 
compatible connecting shapes are determined. The one that 
connects the most neighbor cells to a sound roof structure is 
then chosen and its parameters determined from the parameters 
of the neighbor cells. 
2.5 Manuel Editing 
Because not all roof structures can be fully automatically 
reconstructed, there is a need for manual editing. In our editing 
tool, the decomposition lines can be copied, added, deleted, 
translated and rotated. The cells’ roof shapes are automatically 
reconstructed after every manual step, so that the operator can 
immediately see the results. Once the cell decomposition fits the 
roofs shape, the cell parameters can be manually adjusted or 
even copied from other cells. If the decomposition produces too 
many small cells, then their number can be decreased by a 
merging operation. 
Even though editing the building models using decomposition 
lines is not so straight-forward, we found that operators got 
used to it very quickly and can efficiently produce even 
landmarks with complex geometry. The manual mode also 
allows for more complex roof shapes like mansard, cupola, 
barrel and even some detail elements like dormers. 
3. PROJECTS 
While still in development, we started using the reconstruction 
software in a real production environment. Several large-area 
projects have since been successfully completed. The feedback 
in the early stages of development helped us to recognize and 
adapt to arising problems. Two of our early projects were the 
3D reconstruction of East Berlin and Cologne, two major cities 
in Germany. The 3D city model of Berlin is also available 
online for use in Google Earth (Berlin 3D, 2009). 
3.1 East Berlin, Germany 
The first project with our new software was to perform a 3D 
building reconstruction from Berlin’s LIDAR data. The total 
area of the project was 498 km 2 with approximately 244,000 
buildings. The project was an extension of the original 3D City 
model of Berlin, Germany, which is to date still the largest city 
model transported to the Google Earth platform. Input data 
included a DTM, airborne LIDAR and building footprints. See 
Figure 12 and Figure 13 for the resulting model. 
Due to the large number of buildings in East Berlin and project 
time constraints, photogrammetric extraction was immediately 
deemed as being too time consuming and costly. It was 
therefore decided to use LIDAR data instead. All LOD 2 
building models are geo-referenced geometry, which were later 
textured using aerial oblique imagery. 
The Berliner Roof- a particularly unusual roof type typically 
found on many buildings in Berlin - presented a challenge as 
well as numerous inner courtyards presented problems during 
extraction. Therefore, the reconstruction approach had to be 
adapted to automatically detect this unique roof structure. As a 
result, a total of 17 individual roof types have been additionally 
integrated into the software to enable greater accuracy during 
reconstruction and to reduce the amount of manual editing 
needed. 
As the software was constantly improved during the duration of 
the project, the amount of manual editing needed for the 
reconstructed 3D buildings was reduced from 30 percent in 
denser areas to 20 percent; manual editing for the outer lying 
areas also experienced a sharp improvement: from 20 percent to 
15 percent. 
Figure 12. 3D city model of East Berlin. 
Figure 13. 3D city model of Berlin textured from oblique 
images showing part of the prominent Kurfurstendamm. 
3.2 Cologne, Germany 
The existing 3D city model of Cologne is used by several 
administrative departments as a complement to the existing GIS 
data inventory held by the Cologne Survey Department. It was 
created from the basis of building storeys using two- 
dimensional footprints; therefore the true heights of the 
buildings were not accurate. In order to produce a more realistic 
representation of Cologne in 3D to be used for urban planning 
and emergency response, the survey department decided to use 
the data from the most recent LIDAR flyover to perform a real 
3D building reconstruction. See Figure 14 and Figure 15 for the 
results. 
Cologne’s city boundaries encompass approximately 415 km 2 
with 280,000 buildings; therefore it was decided to use airborne