[TON
> number
her qual-
ive to be
ed. As an
ly is not
edestrian
'equently
d the ter-
an be in-
em based
f a single
if the re-
ese tools
he build-
al part of
DAR.
un
from ter-
> genera-
c visuali-
f texture
)se-range
\R meas-
tual city
process-
are inte-
iming on
cen made
n of data
this con-
of exist-
y a GUI,
ng primi-
espective
ynsuming
erably, if
magery is
is based
areas can
7. Thus, a
siderable
ions have
er of im-
n signifi-
ual inter-
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
action is still necessary. For geometric processing of the pano-
ramic scenes, the exterior orientation is determined from control
points, which are measured manually. Control point information
is provided from the available building models. As it will be
demonstrated in section 3 this georeferencing process can be
sped up, if these correspondences between image and building
model are provided automatically. This can be realised by a
shape based matching of standard terrestrial images to the ex-
isting building models. Additionally, this section demonstrates
the automatic selection of optimal texture for the respective
parts of the building if multiple images have been captured for a
single building. The potential of integrating existing building
models in the processing of terrestrial LIDAR will de discussed
briefly in the final part of the paper.
2. TEXTURE MAPPING FROM PANORAMIC IMAGES
One general problem for the provision of facade texture from
terrestrial images is the large amount of terrestrial scenes to be
processed. As an example, within the project “World Models
for Mobile Context-Aware Systems" (Stuttgart University 2003)
a detailed virtual landscape model of the city of Stuttgart had to
be made available. Within the first phase of the project, facade
texture was collected for a number of buildings in the historical
central area of the city by manual mapping. This manual map-
ping was based on approximately 5,000 terrestrial images col-
lected by a standard digital camera. In order to extract the fa-
cade structures from these images, which were available for
approximately 500 buildings several man months were required.
One approach to increase the efficiency of this process is to re-
duce the number of scenes to be processed for texture mapping
by the application of panoramic images. Frequently, these pano-
ramic images are used for tourist purposes in internet applica-
tions. For this type of application, the scenes can be generated
by the combination of several overlapping images from simple
digital cameras. In contrast, for our investigations a high resolu-
tion CCD line scanner, which is mounted to a turntable was ap-
plied. By these means high quality panoramas can be captured,
which allow for precise photogrammetric processing, i.e. in ap-
plications like 3D reconstruction or mobile mapping. Espe-
cially, if the panoramic scene is collected from an elevated point
in order to minimize occlusion, a large area can be covered.
Thus, texture for a considerable number of buildings is avail-
able from a single scene. In addition to the reduced amount of
images to be processed, the application of panoramic scenes al-
lows to minimize changes in illumination, since image collec-
tion at different epochs is avoided.
2.1 Data Collection
For our investigations, panoramic scenes were captured by the
camera system EYSCAN. An exemplary scene collected from
this system is depicted in Figure 1. In order to demonstrate the
available resolution for this type of imagery, an enlarged section
of the complete scene is additionally presented.
Figure 2 depicts a 3D view generated from the 3D city model of
Stuttgart, which was used as basic dataset within our investiga-
tions. The model was collected on behalf of the City Surveying
Office of Stuttgart semi-automatically by photogrammetric ste-
reo measurement from images at 1:10000 scale (Wolf 1999).
For data collection, the outline of the buildings from the public
Automated Real Estate Map (ALK) was additionally used.
Thus, a horizontal accuracy in the centimetre level as well as a
relatively large amount of detail could be achieved. The result-
565
ing model contains the geometry of 36,000 buildings. In addi-
tion to the majority of relatively simple buildings in the sub-
urbs, some prominent historic buildings in the city center are
represented in detail by more than 1000 triangles, each. The
panoramic image depicted in Figure 1 was collected from the
top of the building, which is represented by the black wire-
frame lines in Figure 2.
Figure 3: EYSCAN camera with cylindrical coordinate system.
