A Field Report on the Modeling of 3D Landmarks in Scholar Projects

Frank Steinicke∗ Tom Feller†

Visualization and Computer Graphics (VisCG) Research Group,

Department of Computer Science, Westfälische Wilhelms-Universität Münster,
Einsteinstraße 62, 48149 Münster, Germany

A BSTRACT in many countries this input data source is available area-widely.

General interest in visualizations of digital 3D cityscapes is grow- When using this data, outlines of building footprints can be placed
ing rapidly, and several applications are already available that dis- precisely into a virtual model. The outlines are extruded to real
play such models very realistically. In order to generate a virtual world heights determined by the GIS data. The described data
3D city model different approaches exist which use miscellaneous sources are usually not used on their own but in combination, be-
input data sources, for example, 2D maps, aerial images, or laser- cause hybrid procedures can yield much more accurate results. Fa-
scanned data, and produce models automatically. However, 3D cades, for instance, can be specified by means of 2D textures based
landmarks which denote highly complex and architecturally promi- on heuristic assignment approaches [3].
nent buildings, e. g., churches or castles, cannot be reproduced in an Despite the benefits of 2D ground maps as an extremely high-
adequate manner by automatic reconstruction. Therefore, architec- level information source reconstruction of virtual 3D city models
tural offices or 3D design companies usually model these entities pose further problems. For instance, while the outline of each build-
manually. In recent years user interfaces of 3D modeling applica- ing footprint is provided with an accuracy in the area of decime-
tions have evolved in such a way that these applications are widely ters automatic reconstruction of virtual buildings – even if the pro-
accepted and easy-to-use – even for non-experts. cess relies on highly detailed sources – lacks certain characteristics
In this paper we present a field report on the manual modeling which cannot be targeted sufficiently in an automatic way. This
of 3D landmarks in which two classes from different schools have issue is illustrated in Figure 1. Figure 1 (a) shows a reconstruc-
participated. We have performed two scholar projects with eight tion based on the proceeding described above. As it can be seen
and ninth grades students. In each project students have chosen despite the accurate outline of the building footprints neither ge-
certain 3D landmarks subject to be modeled. We present the results ometry details nor the overall impression of this building can be
as well as experiences that we have made during this project. obtained virtually (compare to Figure 1 (b)). Hence manually treat-
ment of architecturally prominent buildings is required. In cooper-
Keywords: City visualization, 3D modeling, 3D landmarks ation with two schools we have performed a 3D modeling scholar
project where overall 40 students were involved. Each student was
1 I NTRODUCTION
in charge for one particular building s/he had to model manually.
Three-dimensional geographic visualization environments are of
major importance when exploring a virtual cityscape or when deci-
sions in terms of urban management have to be made [1]. Procedu- 2 P ROJECT D ESCRIPTION
rally modeling virtual 3D cities has been in the focus of computer We have cooperated with two computer science courses from the
graphics research for recent years. While results have consistently eight and ninth grade respectively. Within the class schedule the
improved different approaches have been favored. Basically there teachers have integrated an excursion about 3D modeling in which
are two main approaches in generating a virtual three-dimensional basic facts about geometry and texturing as well as the trade-off
city environment by means of fully-automatic or semi-automatic between complexity and visual representation have been discussed.
procedures. With respect to the approaches different input data After an initial meeting with the students, the cooperations last one
sources are commonly used. The most popular data source for an month, and the students modeled at least two hours per week during
automatic reconstruction of a virtual cityscape are aerial images as their courses.
used for example for photogrammetry [1]. Considerable problems
arise particularly in dense urban contexts when analyzing and inter-
preting this low level input sources. However, in conjunction with 2.1 Tasks and Requirements
other material such as height models or multiple overlap, many in-
Each student was assigned to one particular building, which s/he
formation can be extracted from aerial images such as roof types.
was responsible for during the entire modeling process. The land-
Newly a popular data source for automatic city generation is laser-
marks had to be modeled up to a level-of-detail (LoD) 4, which
scanned data, which in contrast to aerial images represents the ge-
extends block models to textured facades, detailed roof types and
ometry of the surface directly. But this procedure involves almost
modeled interior. The students had to review the real buildings
prohibitive costs, and furthermore it may suffer of low density of
on site in order to record notes and to take images from different
measured points. Another very popular method for automatic re-
perspectives. Thereupon, they reconstructed the buildings virtually
construction relies on two-dimensional ground maps or plans from
as close as possible to the real world while providing a sufficient
Geographic Information Systems (GIS), which contain the outline
trade-off between their complexity and visual representation. As
of building footprints, their heights, streets, street furniture and fur-
additional motivation an internal jury has judged the models and
ther information. Since data acquisition can be very cost-effective
the winners have received a free flight over the real city.
∗ e-mail: [email protected]
† e-mail:[email protected]
2.2 Tools
During the modeling process the students were provided with cer-
tain data sources and tools. At the beginning of the project students
have been made acquainted with these media.
 (a) Automatic reconstruction (b) Real building (c) 3D model in Cinema4D (d) 3D model in CiVisE

Figure 1: Illustration of the reconstruction process: (a) an automatically reconstructed building and (b) its a real counterpart (churches in 2D
ground maps are referenced as one-storied buildings). (c) during the manual reconstruction process in a 3D modeling application details can be
considered and (d) the model can be integrated in the city visualization environment.

2.2.1 Virtual Blockmodels those buildings modeled by university students, even if less time
In order to accelerate the modeling process, but also to ensure qual- was available. The high average number of triangles used for each
ity assurance, we supported the students by means of given basis building gives an indication about the quality of the models. The
data. We exported a 3D file of each target building, which contains file volume size averages to approx. 450KB, which indicates on the
the vertices of the automatically generated block model. We ex- hand side a sufficient quality of geometry as well as textures, but
ported this block model in the 3DS data file format without textures on the other hand it allows real-time visualization within our city
which corresponds to a LoD 0. These 3D models can be imported visualization environment (CiVisE) applications [3]. The usage of
in arbitrary 3D modeling applications such as Maya 3D, AutoCAD, SketchUp was definitely advantageous since the students learned
Cinema4D, SketchUp etc. (see Figure 1 (c)). very fast to work with this application. Moreover, their ability to
use the software exceeds the skills of their teacher already after a
2.2.2 Image Capturing and Processing short period of time. Some of the students still model landmarks
for the Google Earth interface.
Before the student could start the modeling process they had to re-
After the project we have performed a survey with the students.
view the real buildings in order to reveal certain characteristics and
In their subjective comments they have revealed that they have
details which are not available in the block model. In particular the
learned a lot about modeling and that they have got interesting in-
roof type as well as a realistic facades are essential properties of
sights into computer graphics and virtual environments. In particu-
landmarks. For this purpose the students captured digital images
lar they have enjoyed the final presentation of their work in our vi-
of the buildings from different perspectives. These textures have
sualization laboratory, i. e., a passive stereoscopic back-projection
been mapped to each building wall of the block model and have
environment with head tracking [3]. Thereby, by means of virtual
been exploited as basis for the facades’ geometry. Thus the stu-
flights and walks they were able to inspect their modeled buildings
dents could easily model windows, doors, indentation, pillars etc.
in an immersive way. Some of the students have already applied for
by means of their positions and measures projected on the textures.
an internship in our research group.
Although the images have been captured early in the morning some
disturbing objects occluded the facades, e. g., trees, vehicles or hu- 4 C ONCLUSION
mans. Hence all textures have been post-processed using ordinary
2D graphics applications in order to diminish such elements in the In this report we presented a field study about cooperations between
textures. two schools and our university which was focused on modeling of
3D landmarks for CiVisE. The cooperation was very successful in-
2.2.3 3D Modeling Software dicated by the delivered virtual 3D models as well as the experi-
ences and facts students have learned during the project. The local
The students were not restricted to certain modeling software, how-
press has reported about the cooperation in some articles and the
ever, most of them used Google’s SketchUp. Due to the easy-
models of the students were subjects of some commercial demon-
to-use interface and the short required training time working with
stration videos we have prepared to certain companies and orga-
SketchUp emerged as a very efficient procedure. After a two hours
nizations. The work inspires us to continue working with schools
course where basic features have been explained the students were
even in other visualization domains, for instance medical visual-
able to begin to model their buildings. Two teachers were in charge
ization. Currently, we have developed techniques which alllow to
with the classes and supported the students when any problems oc-
export the entire city model to Google Earth’s KML file format.
cur. In Figure 1 (c) the virtual building corresponding to the real
counterpart shown in Figure 1 (b) is shown in a 3D modeler When R EFERENCES
a 3D model has been finished the 3D file can be imported into the
[1] M. Beck. Real-Time Visualization of big 3D City Models. Interna-
virtual city visualization environment (CiVisE) and it can be ex-
tional Archives of the Photogrammetry, Remote Sensing and Spatial
changed with the building that was generated automatically before.
Information Sciences, XXXIV(5/W10), 2003.
Due to this manual approach all desired details can be incorporated [2] J. Döllner, H. Buchholz, F. Brodersen, T. Glander, S. Jütterschenke, and
into the 3D landmark model [2]. A. Klimetschek. SmartBuildings – A Concept for Ad-Hoc Creation and
Refinement of 3D Building Models. In Proceedings of the 1st Interna-
3 R ESULTS AND E XPERIENCES tional Workshop on Next Generation 3D City Models, 2005.
Overall we evaluate the cooperations with both schools as very suc- [3] F. Steinicke, T. Ropinski, K. Hinrichs, and J. Mensmann. Urban City
cessfull. The students were highly motivated when modeling con- Planning in Semi-immersive Virtual Reality. In Proceedings of the
tent that was subject to be integrated into a real-world application International Conference on Computer Graphics Theory and Applica-
used for several demonstrations. The quality of the modeled build- tions (GRAPP2006), pages 192–199. INSTICC Press, 2006.
ings are comparable and moreover do often exceed the quality of