A different semi-automatic approach for building extraction 
and photo texture deduction from aerial images is shown in 
[LLS95] and other publications from the University of Bonn. 
1.3 Modeling 
Photo textured 3D CAD models (see figure 1) have so far 
been created by hand at great expense [GMB95]. Much work 
has to be accomplished manually [GMB95] in the generation 
process of a photo-realistic 3D CAD model. Transferring GIS 
data results in building-boxes, which are rendered with tex- 
tures from aerial photography and images taken from street 
level. The fusion process requires extensive manual work for 
the modeling, thus, a high degree of detail (bay-windows, 
balconies,. . .) is only being achieved in experimental and small 
models. 
1.4 Motivation 
According to the previous discussion, the main objective of 
ongoing work is to find methods for semiautomatic or au- 
tomated data acquisition and object reconstruction. This is 
valid for reconstruction from both aerial photographs and ter- 
restrial imagery. 
The support of an intelligent fusion of input data, the au- 
tomatic recording of high quality photo texture of building 
facades and at least tools for the semiautomatic creation of 
geometric models of facades have to be provided. 
We will focus on the design of a system for the automatic 
recording of building facades in this report. 
2 ISSUES OF THE DATA ACQUISITION 
2.1 Principal recording configuration 
The principal design of the system is shown in figure 2. A 
vehicle based recording unit is moved laterally along a build- 
ing facade. The 3 vertical CCD-lines are recording the facade 
continuously and simultaneously under different angles in or- 
der to allow stereo reconstruction and occlusion elimination 
in a subsequent processing step. 
  
Figure 2: Principal recording configuration: The as- 
sumed coordinate system, the line sensors and the area 
sensor are shown. The car's center of mass is indicated 
by m and the vertical outrigger is denoted by or. 
An additional area sensor is indicated in figure 2 to support 
the automatic orientation and camera tracking process. Odo- 
metric sensors and a laser distance meter support an algo- 
rithm for the reconstruction of the recording path. 
For a more detailed description of the complete implementa- 
tion see [Mar95b] or [MSh96]. 
AN 
10m A. 
  
| Mp 
  
Figure 3: Recording constraints in urban environment 
in practice: One distance corresponds to the old part 
of town, and the other to the area around the center 
of a city with wider streets. 
Two examples of recording constraints faced in urban envi- 
ronment are shown in figure 3. One distance corresponds to 
a narrow older part of a town, and the other to areas around 
the center of a city with wider streets. 
The distance between the optical center and the facade usu- 
ally is limited by the narrowness of streets and the height of 
the outrigger is limited by the overhead contact line of street 
cars and an unacceptable amount of oscillation. 
  
  
  
Figure 4: CCD platform: The relative orientation of 
3 linear CCD arrays in one con-focal plane is shown. 
The shift of the center of the lines from the optical 
center also can be observed. The following angles were 
chosen: $o — 20?, ko ~ 10°. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B1. Vienna 1996 
  
The s 
[Hof8: 
metric 
ence t 
smoót 
ture, | 
path : 
tical à 
The s 
tical e 
figure 
advan 
2.2 
Ther 
fined | 
in pra 
orient 
qualit 
  
Some 
systen 
conce 
An ex 
lems r 
The c 
quiren 
pleme 
speed 
far be