Generating precise and accurate 3D city models using photogrammetric data

Full text

GENERATING PRECISE AND ACCURATE 3D CITY MODELS USING
PHOTOGRAMMETRIC DATA
O.Emem*, F.Batuk,
YTU, Division of Photogrammetry and Remote Sensing, Beikta Istanbul, Turkey- (oemem, batuk)@yildiz.edu.tr
Comission IV, WG IV/6
KEY WORDS: Photogrammetry, Three-dimensional, Modeling, Surface, Visualization, Rendering, Virtual Reality
ABSTRACT:
This paper presents a study, which aims to generate 3D city models from existing 2,5D maps. A 3D city model is generated via CAD
software using the existing photogrammetric 2,5D maps of a chosen pilot area. 3940 buildings, 8000 trees, 810 streetlight lamps are
modeled in the scope of the project. Two 1/5000 scaled orthophotos are used for visualization of the surroundings. In order to reach
the high vertical accuracy for buildings, extra editing processes are applied to the all buildings in the project area. Therefore, the
vertical errors of buildings were corrected on 2.5D photogrammetric maps. Finally a precise and an accurate 3D city model of
Beikta Region is generated and converted into VRML. Using VRML format, it is also possible to visualize and animate the 3D city
model on the internet.
1. INTRODUCTION
Developments in computer technologies supplies us new
tools, which can be used for administration, planning,
applications and making things easier in daily life.
Furthermore these tools are used to increase the quality of
daily life and to save the nature. The names of these tools are
CAD, GIS, GPS, laser scanning, radar, digital
photogrammetry, 3D city modeling from the point of view of
measuring, evaluation and presentation. These tools have
been developed since 1960’s Instead of using the hard-copy
maps, new technologies such as visualizing the
improvements of city on the display and having a tour on a
3D virtual city make things easier for architects and city
administrators, and these abilities let them decide effectively
and consistently.
While the applications in digital environment were
investigations of possession, taxing etc. in 80s; those
applications were moved to planning and preserving habitat
in 90s. Nowadays it is possible take a tour around cities in
virtual reality models, and those models give opportunity to
make measurements and to design new concepts.
The 3D city models, which generated by different methods
can be transferred into GIS easily and used as a base for GIS.
In addition, it is possible also to make queries on the 3D city
models, and these 3D city models can be published on
internet for civilian users (Zlatanova, 2000).
Nowadays, 3D city models are generated using 3 main
approaches: automatic, semi-automatic and manual. Thus,
existing 2D and 2.5D maps are bases for semi-automatic and
manual 3D city model generation.
Photogrammetry is an appropriate tool to provide information
about man-made objects, vegetation cover and the
like(Gruen, 1999). Photogrammetry and digital image
processing techniques play important role to make
visualization technology practical and cost effective. In the
3D city modeling, simulation is done by using true
information via photogrammetry. Photogrammetry appears to
provide the only economic means to acquire truly 3D city
data (Förstner, 1999). By means of true information in the
content of the precision and accuracy, the 3D city models can
be used for simulation and analysis from visualization and
animation (Emem, 2002b).
In spite of high vertical and horizontal accuracy of
photogrammetric maps, it is not possible to use them directly
in the 3D city modeling. Especially the buildings on the
digital maps have some vertical errors, when we examine
them from the perspective view. Hence, these perspective
errors should be edited before using them in the 3D city
models (Emem, 2002a).
In this paper, a manual 3D city model generation using
existing 2.5D maps project is presented. But this paper
mainly focuses on editing the height differences on buildings.
2. METHODOLOGY
In this study, some certain methods are applied for the 3D
model generation as follows:
• Examining of 2.5D maps for elimination of the data
which is not used
• Generation of DTM
• Editing the vertical differences of the buildings and
solid modeling of buildings
• Preparation of the 3D cell libraries and locating
them on DTM
• Visualization
• Virtual reality on the web.
The modelings of roof of the buildings are not aim of this
study. Because, modeling of the roofs requires extra data,
which does not exist in the 2.5D photogrammetric maps.
3. DATA FOR THE 3D CITY MODELS
The 3D city models contain data, which is related to the
spatial objects. Those objects might be located over the
surface, on or under the surface, and those are includes the
details, which has certain shape and location. Those objects

are generally topography, buildings, transportation, plant
cover, infrastructure and other details related to the goal of
application.
Actually, the 3D city models are focused on buildings;
however it may contain roads, green cover, trees etc.
according to the usage of model.
The data is most essential component of the 3D city models.
Therefore, data should be in appropriate form and accuracy.
If the data was collected as a purpose of generation of the 3D
model, it will largely be suitable for use. But, when we
consider using the existing data, the extra editing jobs will be
inevitable.
The 3D city models are highly useful for city administration
however; collecting required data is costly and very hard.
Therefore, using existing data for model generation is
cheaper and practical. Especially existing maps produced by
phogrammetry is much more suitable for 3D city model
generation (see Figure 1, 2).
Figure 1. Top view of digital photogrammetric map
Figure 2. Perspective view of digital map
Addition to the maps, orthophotos and rectified image of the
building facades are used for visualization of the 3D model.
These images are draped and rendered on surfaces for photo-
realistic view.
4. THE DATA PREPARATION
In this study, 1/1000-scaled digital maps are used. In order to
reduce the mess, some unnecessary data such as
administrative boundaries, walls, railways etc. is eliminated
from the maps. Selected 53 object types are selected and
used. Some of those objects are as follows:
• DTM points
• Breaklines
• Sea and lake boundaries
• Sea ports
• Buildings
• Roads
• Plant cover boundaries
• Trees and lamps
After elimination, those data are classified according to its
type (see Figure 3). For instance all the buildings types such
as residential building, mosque, factory etc. are classified in
the same level. This class is used for generation of the solid
models of the buildings. Some certain classes are constituted
for all data types.
Figure 3. Buildings
4.1 Generation of DTM
In order to generate DTM, related object’s level is used. This
level contains features and height points, which can define
the topography of the area. These features are DTM points,
declivities, contour lines, boundaries etc. All these features
are converted to the points. Finally terrain was triangulated
with these points. But some of these points have wrong z
values and needs to be eliminated. During the production of
the maps, human made mistakes causes this wrong z values.
Sometimes operator can measure the heights incorrectly.
Therefore these values should be edited or deleted from the
data during the process.
In order to correct them, manual or automatic procedures can
be applied. In manual way, this problem is solved interactive
control and editing (Yastikli, 2002). Operator triangulates the
surface and examines the triangles visually. But this method
causes neglect of the lower errors and time consuming.
Automatic methods are editing the points automatically with
selected parameters (Yastikli, 2002). But using this method
can cause loosing the small details on the surface. In this
project both of these methods are used for best result. Thus,
height errors and faulty triangles are corrected or deleted in
order to reach the precise DTM (see Figure 4). Triangulated

surface is converted to 250.000 sampling points and
grid/lattice (see Figure 5).
Figure 4. Triangulated surface
Figure 5. Lattice of the surface
4.2 Editing The Vertical Differences of The Buildings and
Solid Modeling the Buildings
Like the z errors in the point clouds, the buildings have
height errors in it as well. The reason of these errors is again
arises during stereo compilation like other z errors. Operators
sometimes see different points other then the roof corners of
the building (Figure 6). Even if they compile it with true z
value, it appears as an error when seen from the perspective.
For instance he/she can set the floating mark on the surface
instead of roof. They can not be realized from the top view
(see Figure 6). So that errors can not be edited during 2D
editing phase. This kind of errors is very rare, but can be
encountered and needs high attention (Emem, 2002a).
Figure 6. Wrong point compilation error
The other error arises because of the roof corner height
differences. Despite the operator compiles the building
correctly, this error comes from the nature of the view and it
is inevitable. Normally the height of the building corners
should have the same height value. But some buildings which
have triangle roof types are compiled with different height
values (see Figure 7).
Figure 7. Triangle roof
This kind of buildings can not be realized like the preceding
error during the 2D editing phase (Figure 8).
Figure 8. Height difference of the roof
Even if both errors are not critical for conventional usage of
maps, they are important errors for 3D city modeling.
Therefore these errors should be corrected before solid
modeling the buildings.
If we consider the compilation process of buildings, the
operator starts from a corner of the upper point of the
building. Therefore, there are 2 ways to correct height
differences on the buildings. First, accepting the arithmetical
average value of the building data points. But this assumption
causes changes on the values from the real values. Other way
is to assume the correct value as the first compilation data
point the building. Because, operators generally fix the
parallax on the first data point correctly and more carefully in
addition the most of them start to compilation from upper
point of the building instead of the roof corner (Emem,
2002a). In this study, the first data point of the building is
assumed as the correct value. Therefore all the buildings on
the project area were edited with the accepted height values
(see Figure 9, 10). Some buildings on the project area such as
stadium and complex shaped hotel were edited separately
from others.
Figure 9. Correcting values according to accepted value

After editing, the entire buildings are solid modeled by
extruding them down to the DTM (see Figure 11, 12, 13).
Figure 10. Before editing and after editing
Figure 11. Extruding the building
This height correction process provides buildings to be better
shaped for the 3D city models. Since it is hard and expensive
process to extract roofs from the existing maps, the roofs of
the buildings are neglected. Editing of the height differences
on the buildings are done automatically.
Figure 12. Solid models of buildings
4.3 Preparation of the 3D cell libraries and locating them
on DTM
In this study, a symbology library is prepared for two kinds
of trees and street lamps. Those symbologies are converted to
the cells to use them in the 3D city model. The heights of the
trees and lamps are assumed as same values (see Figure 13,
14, 15). Therefore, these 3D cells are placed just for
visualization of the 3D city model.
Figure 13. Pine tree symbol
Figure 14. Symbol for other tree species
The trees on the existing map, which is used for this study
contains two species of tree. These are pine trees and other
species. Therefore, two types of tree symbols are used in this
study. One symbol is designed for street lamps (Figure 15).
Figure 15. Street lamp symbol
However, the trees and the lamps are displayed as planar
objects in the digital maps, they contain height values. They
are compiled from the intersection point with the surface if
possible. In the 3D city model, those 3D cells are placed
instead of the planar objects on the map.
4.4 Visualization
In this study, two different visualized city models are
generated. First, the orthophotos of the project area are used.
Orthophotos are draped on the DTM, thus photo-realistic
view of the surface is generated (see Figure 16).

Figure 16. Photo-realistic view of the surface
After draping solid modeled buildings are merged with this
view (Figure 17).
Figure 17. 3D city model
Secondly, the 3D cells and colored DTM is used. Primarily
green cover areas and watery areas are determined and these
areas are painted green and blue (Figure 18).
Figure 18. Colored DTM
Then, 3D cells are placed and merged with the solid
buildings. Additionally roads and green covers areas are
merged with the other files. Finally, two different visualized
3D city models are generated (see Figure 19,20).
Figure 19. 3D city model with 3D cells
Figure 20. 3D city model
4.4 Virtual Reality
After generation of the 3D city model, it is aimed to submit
the model to the citizens on internet (Özmutlu, 2001). Thus,
VRML is chosen to achieve this purpose (Zlatanova, 2000).
Therefore, the 3D city model is converted to VRM language.
MicroStation Software is used for conversion. Finally, the 3D
city model is published as VRML via web page (see Figure
21).
Figure 21. VRML view
5. CONCLUSION
Developments in computer technologies have improved new
tools, which can be used for administration, planning,
applications and making things easier in daily life.
Furthermore these tools are used to increase the quality and
to save the nature.

Photogrammetry is economical way to provide true height
values, which is required for 3D city modeling. Since,
collecting the data is costly and very hard, 2D and 2.5D
photogrammetric maps are used as base. Therefore, accuracy
of the city model is based on the accuracy of the base map.
Since existing maps are not containing data about the roofs of
the building, it is recommended to gather required data
during production of the maps and to take this as a common
rule. Otherwise same procedure will be done repeatedly and
data will be duplicated.
Because, the collection of the data required for the 3D city
models is very costly, using existing data is became current
issue. But using existing data directly in 3D model generation
might be full of errors. Especially buildings, which are the
main objects of the 3D city models, have distinct errors.
Using the data without editing will cause X, Y and Z errors in
the 3D city models. Additionally, view of the model will be
different from the real world shapes.
In order to generate accurate and precise 3D city models from
the existing photogrammetric maps, the data should be edited
as required.
REFERENCES
Emem, O., 2002a. 3D Modelling, Design and Application.
MSc. Thesis. (in Turkish), Yildiz Technical University,
Istanbul
Emem, O., Yastikli, N., Balik, F., Alkis, Z., 2002b. Creating
3D Photo Realistic Models for Visualization of Historical
Buildings. 3rd International Symposium Remote Sensing of
Urban Areas, Istanbul
Förstner, W., 1999. 3D-City Models: Automatic and
Semiautomatic Acquisition Methods, Photogrammetric
Week, Wichmann, Karlsruhe
Grün, A., Wang, X., 1999. CyberCity Modeler, a tool for
interactive 3-D city model generation. Photogrammetric
Week, Wichmann, Karlsruhe
Özmutlu, ., 2001. 3D Geoscientific Modelling, 3D
modelling, Principles. Applications Course Note, 3D Model
Building & VRML, p.p 1-16.
Yastikli, N., Jacobsen, K., 2003. Automatic Digital Elevation
Model Generation, Problems and Restrictions in Urban
Areas. Journal of Yıldız Technical University, Vol. 2003-2
pp.38-46
Zlatanova, S., Tempfli, K., 2000. Modelling for 3D GIS:
Spatial Analysis and Visualization Through The Web.
IAPRS, Vol. XXXIII, Amsterdam