ON THE REFINMENT OF URBAN MODELS BY TERRESTRIAL DATA COLLECTION
Norbert Haala
Institute for Photogrammetry (ifp), University of Stuttgart, Germany
Geschwister-Scholl-Strasse 24D, D-70174 Stuttgart
Norbert.Haala@ifp.uni-stuttgart.de
Commission III WG 7
KEY WORDS: Visualisation, Virtual Reality, Photo-realism, Terrestrial
ABSTRACT:
Urban modets can be collected area covering and efficiently based on aerial data like stereo images or LIDAR. While a large number
of applications like simulations are already feasible based on the available three-dimensional building representations, a further qual-
ity improvement is required for some other tasks. Especially, if very realistic visualisations from pedestrian viewpoints have to be
generated, the quality and amount of detail, which is available for urban models generated from aerial data has to be improved. As an
example, due to the viewpoint restrictions of airborne platforms, image texture for the facades of the buildings frequently is not
available from airborne sensors. Since this information is crucial to improve the visual appearance of urban models for pedestrian
perspectives, alternative data sources have to be applied. Thus, image texture for the faces of existing building models is frequently
generated based on manual mapping of terrestrial images. For this purpose, correspondences between the existing model and the ter-
restrial image have to be provided. As it will be discussed within the paper, the efficiency of the manual mapping process can be in-
creased considerably by the application of panoramic scenes. If these panoramic scenes are collected from a high-level system based
on a rotating CCD line scanner, large areas are covered at high resolution and superb image quality. Thus, the processing of a single
scene is sufficient to extract texture for a considerable number of buildings. Texture mapping can additionally be sped up if the re-
quired correspondences between image and model are provided from automatic tools. If multiple images are collected, these tools
should also allow for an automatic selection of the image, which is most suitable for texture mapping of a specific part of the build-
ing. While the main emphasis of the paper is on the refinement of existing urban models based on terrestrial images, the final part of
the paper will briefly discuss the benefits of integrating existing building models to the processing of data from terrestrial LIDAR.
for pedestrian viewpoints. This is only feasible, if data from ter-
1. INTRODUCTION restrial platforms is additionally avaialable during the genera-
tion of the urban models. As an example, photorealistic visuali-
The development of tools for the efficient and area covering sations of building façades require the availability of texture
collection of 3D city models has been a topic of intense research from terrestrial images. Of course, approaches like close-range
for the past years. A good overview on the current state-of the- photogrammetry or the evaluation of terrestrial LIDAR meas-
art is for example given in (Baltsavias, Grün, van Gool 2001). urement allow for an autonomous collection of virtual city
Meanwhile, a number of algorithms are available, which are models. Nevertheless, the efficiency of terrestrial data process-
usually based on 3D measurements from airborne stereo im- ing can be increased considerably, if existing models are inte-
agery or LIDAR. For this reason, the production of virtual city grated during this evaluation. For this reason, we are aiming on
models is more and more becoming a standard task of photo- ^ (ne refinement of existing urban models, which have been made
grammetric data collection, resulting in a growing availability of ^ available from airborne data sources by the application of data
data sets, which include 3D representations of buildings. from terrestrial platforms.
Frequently, these 3D models are used to generate photorealistic As already discussed, a common task to be solved in this con-
visualisations of the urban environment. Typically, this type of text is the mapping of terrestrial images to the facades of exist-
application is required in the context of urban planning, tourism ing 3D building models. Frequently, this is realised by a GUI,
or entertainment, like games based on real locations. Airborne which allows a human operator to select corresponding primi-
data collection is suitable to efficiently provide a complete set ^ tives between the available building model and the respective
of 3D building models, mainly representing the footprints and ^ images, As it will be discussed in section 2, this time consuming
the roof shapes of all buildings at sufficient detail and accuracy. process of manually mapping can be sped up considerably, if
On the other hand, viewpoint restrictions of airborne platforms panoramic scenes are applied. In our experiments the imagery is
frequently limit the amount of information, which can be made collected from a high-level panoramic camera, which is based
available for the facades of the buildings. For this reason, build- on a rotating CCD line scanner. By these means large areas can
ing models collected based on airborne data can only be applied be covered at high resolution and superb image quality. Thus, a
for high quality visualisations if virtual images from elevated single scene is sufficient to provide texture for a considerable
viewpoints have to be generated. While this type of images is number of buildings, especially if suitable camera stations have
preferable in order to generate overviews of larger areas, a peen selected during image collection. Since the number of im-
number of applications in the context of urban planning or 3D ages to be processed for texture mapping is cut down signifi-
navigation require visualisations at very high degree of realism cantly, the overall effort is reduced, even though manual inter-
564
Intei
actic
rami
poin
IS pi
dem
sped
mod
shap
istin
the:
parts
sing
mod
brief
3.
One
terre
proc
for ^
a det
be m
textu
centi
ping
lecte
cade
appr
One
duce
by th
rami
tions
by tl
digit
tion
pliec
whic
plica
ciall;
in oi
Thus
able
imag
lows
tion.
2.1
For
came
this
avail
ofth
Figu
Stutt
tions
Offic
reo 1
For «
Auto
Thus
relat
