The Dresden Frauenkirche — rebuilding the past

Brian Collins^ Dave WilliamsS Robert Haak^ Martin IVux^, Herbert Herz^, Luc Genevriez',
Pascal Nicot^, Pierrick Brault^, Xavier Coyere', Burkhard Krause'*, Jens Kluckow^ & Armin
Paffenholz^
e 'IBM UK Scientific Centre, ^IBM Deutschland Informationssysteme GmbH, 'ARC (Audiovisuel Realisation
Conseil), ^TransCAT Nord GmbH)

3.1. Historical background begin. Before long the bell-shaped cupola of the
Frauenkirche will tower over the Elbe river and roofs of
For two hundred years, the marvellous bell-shaped cupola Dresden once more.
of the Frauenkirche, or Church of Our Lady, stood above It is a tradition for IBM to make contributions to so-
the roofs of the historic centre of Dresden and dominated cial, educational and cultural activities. IBM supports many
the famous Elbe panorama. Built between 1726 and 1743 projects of public interest under the framework of its Cor-
from a design by the architect George Bahr, the Dresden porate Social Responsibility Program. For the reconstruc-
Frauenkirche was the most important church building of tion of the Frauenkirche, IBM is providing support with
German Protestantism and at the same time one of the great technical equipment, administrative assistance and the or-
masterpieces of European architecture (Fig. 3.1). On 15th ganisation of events in order to increase the awareness of
February 1945, two days after the devastating bombing of potential sponsors. This current project, the computer re-
Dresden, the domed building, constructed entirely of stone, construction, is one major activity within the context of the
became so weakened by the subsequent fire-storm that it actual reconstruction of the Frauenkirche.
collapsed.
Only two fragments of the walls and an enormous heap
of rubble remain (Fig. 3.2). The ruins of the Frauenkirche
3.2. Computer reconstruction
came to symbolise the destruction of Dresden and have stood The use of computer graphics in Archaeology and National
as an admonition for more than forty-five years. However, Heritage has a surprisingly long pedigree. One of the first
the beauty of the baroque Frauenkirche was never forgot- animations was shown as part of a BBC Chronicle pro-
ten. Its rebuilding, planned since the end of the war, was gram which was broadcast on 27 March 1984 and showed
prevented by the economic and political conditions that ex- a reconstruction of Roman Bath. Though the computer re-
isted under communist rule. But the time for the rebuild- construction was done by Woodwark in the early 1980s, the
ing has now come and Dresden, the "Florence on the Elbe", work was not published until 1991 (Woodwark 1991).
will be able to recover its symbol and an important part of Another animation of a computer reconstruction, this time
its former beauty. At the same time, world culture will be of the Saxon Minster in Winchester, was made at the IBM
given back one of its great works of architecture. UK Scientific Centre in the mid 1980s (Burridge et al.
The Frauenkirche is to be rebuilt entirely in its origi- 1989). A comprehensive survey of the field is to be found
nal form, using historic building techniques and the origi- in Reilly (1989) and the references therein.
nal material, sandstone from the Elbe valley. The surviving More recently one of the authors of this paper,
documentation and detailed architectural drawings, pro- Genevriez, has directed animations of a computer re-
duced during an earlier restoration between 1938 and 1943, construction of Paris in 1789, and of an Abbey, that of Cluny
will make this possible. The surviving building fragments in France. The latter animation was presented at the recent
will be incorporated and most of the original stones reused. Imagina '93 conference.
The baroque interior furnishings including the famous or- The following sections describe the technical stages
gan, originally built by Gottfried Silbermann, are also to be which were used to reconstruct the Dresden Frauenkirche.
restored. When completed the Frauenkirche will once again However, this project was not just a technical exercise. To
be used for church services and for concerts. make this kind of animation both serious and credible it
The rebuilding of the Frauenkirche will cost ap- was necessary to work with curators and archaeologists,
proximately 160 million German Marks and will take about and to perform exhaustive research of the available docu-
ten years. In view of the difficult economic problems and mentation. This was crucially important to ensure that what
high costs of redeveloping former East Germany, it will not was reconstructed was not only historically correct but also
be possible to make use of German federal, state, or church conveyed an "atmosphere" and "emotions" that were au-
funds. For this reason, a "Society for the Rebuilding of the thentic.
Frauenkirche in Dresden" has been formed to take on the
enormous task of collecting donations from all over the 3.2.1. Geometric CATIA 3D model
world. The building plans are now under way, the ruins The detailed architectural drawings which were produced,
have been stabilised, and the work of rebuilding will soon and contemporary photographs which were taken during

19
B. COLLINS ET AL.

Figure 3.1: A painting of the Dresden Frauenkirche, from Figure 3.3: A photograph of the Dresden Frauenkirche
the 18th Century. taken before its destruction in 1945.

the "inside" of the building so that, for example, the altar is

included in the "inside" model but is excluded from the
"outside" model. Similarly separate models were produced
for the existing "ruins" and for the whole "reconstructed"
building.
As the Frauenkirche is not a "normal building" with
rectangular walls, beams etc., geometric reconstruction
presents a major problem. The structure of the building is
very complex, especially the inside of the church and the
roof parts. To create these objects, CATIA works with "com-
plex solids" and "advanced surfaces". Only CATIA allowed
us to achieve these results in a period of just five weeks.
One example is the very complex roof of a staircase tower.
This object was constructed with the help of solids and
"Boolean operations". Two prismatic solids were created
using different contours, these were then intersected at an
Figure 3.2: A photograph of the ruins taken after 15th angle of 90 degrees resulting in a "perfect" roof. Such
February 1945. multiple solids were combined into larger objects called
workspaces (for example the cupola) and these were in turn
combined to form the complete building.
The level of detail varied considerably, but closely mir-
the 1938-43 restoration, were the main sources for the gen-
rored the detail available in the drawings and contemporary
eration of the geometric CATIA 3D model (Figs. 3.3 and
photographs. In some cases very fine details were avail-
3.4). The whole model was input manually, but it was split
able, for example the pulpit, but in others only gross details
into several parts to make the project more manageable.
existed, for example the outside stonework. The over-rid-
Thus, separate models were produced for the "outside" and

20
 THE DRESDEN FRAUENKIRCHE — REBUILDING THE PAST

^y

fMfiê^^^aaas^^^

Figure 3.4: An architectural drawing produced for the Figure 3.5: A geometric model of the "outside" using
1938-43 restoration. CATIA.

ing philosophy in the construction of the geometric model

was not to add detail where it was not available from the
drawings or the photographs.
Only geometric information was input to the model
and no information about the colour or texture of particular
components was included. However, false colours were used
to differentiate parts of the model, thus the windows were
coloured blue and the roofs red.
3.2.2. Photo realistic model — 2D
The geometric model was not detailed enough to produce a
photo realistic reconstruction. Therefore, it was necessary
to supplement it with details of the baroque furnishing, the
pulpit and the organ. Where possible scanned images of
photographs from both before and after the 1938-43 resto-
ration were used. For the remaining details photographs
from contemporary churches in Germany and the United
Figure 3.6: A geometric model of the "inside" using
Kingdom (St. Paul's Cathedral) had to be used.
All these images, some black and white, some in col- CATIA.
our but over 60 years old and some in colour from 1992,
had to be colour matched. This was achieved using a col-
our palette editor. The images were also scaled and warped the hemispherical geometric model of the dome. NEFERTTFI
where necessary. Thus the painted inner-wooden dome was was used for these image manipulation stages.
reconstructed by warping the 2-dimensional scanned im-
age into a 3-dimensional image to be texture mapped onto

21
B. COLLINS ET AL.

Figure 3.7: Image of the "reconstruction outside" Figure 3.8: Image of the "reconstruction outside'
rendered using TDImage. rendered using TDImage.

3.2.3. Photo realistic model — 3D to 2 hours to render and a total of 2800 images were pro-
duced during this stage (Figs. 3.7-3.10).
The geometric model was transferred from CATIA to
TDImage which was used for this stage. It was kept as 3.2.6. Post-production
"outside" and "inside" models and as "ruin" and "recon-
struction" models. Details which were missing in the geo- Because the model was split into an "outside" and an "in-
metric model were added where available, based on side" there were two periods in the animated sequence, (on
photographs of the original church or of contemporary entering and exiting through the main door), when the cam-
churches, in particular the organ, the pulpit, and the gal- era needed to "see" details from both the "outside" and "in-
leries. The original colours were added where known and side" models. Thus if only the "outside" model were used
texture mapping and bump mapping, based on the 2-di- and the main door was open there would be no details of
mensional scanned images were extensively used for the the "inside" visible through the door opening. Similarly if
reconstruction of the baroque furnishings. Finally, appro- only the "inside" model were used and the main door was
priate illumination, both for the "outside" and "inside" open there would be no details of the "outside" visible around
models was chosen. the door opening. To overcome this two sequences were
rendered using the same camera trajectory, one using the
3.2.4. Animation preview "outside" model and one using the "inside" model. As well
as the 24 bits of colour for each pixel in the resulting ren-
For the animated sequence the camera ttajectory was de-
dered images there were also mask bits. These bits were
rived by choosing the camera position for six key frames
subsequently used in a compositing stage to decide whether
and using TDImage to interpolate the camera positions for
a pixel was used from the "outside" or from the "inside"
the intervening frames, typically about 500 between pairs
rendered image was used in the final composited image.
of key frames. The complete camera trajectory from
The lettering in the titles, explanatory text and credits
TDImage and the geometric model from CATIA were both
were all generated as rendered images and subsequently
transferred to Data Explorer. A ray-cast rendered anima-
composited with appropriate still frames from the animated
tion sequence was then generated at about 1 frame per sec-
sequences. TDImage was used both for the compositing
ond and the animation checked by replaying at the planned
and for producing the titles, explanatory text and credits.
rate of 25 frames per second on an IBM POWER Visualization
Where artistically appropriate "fades" were used for
System (see below). This approach was substantially faster
the transitions between pairs of title, explanatory text and
than small-size ray-traced rendering of each frame which
credit images and between some animation sequences.
is the more traditional method. Based on these animation
These were typically for time periods of one to two seconds
previews the photo realistic model and camera trajectory
(25 to 50 frames). This was achieved by blending an in-
were refined in an iterative process.
creasing percentage of the final image with a correspond-
3.2.5. Rendering ing decreasing percentage of the initial image. Data
Explorer was used to generate these fade sequences.
Once the camera trajectory and the photo realistic model The final animated sequence contained nearly 3000
were finished the final animation sequences were gener- frames with an additional 32 still frames for tides, explana-
ated by ray-traced rendering using TDImage. This used tory texts and credits. The full length computer reconstruc-
both the "outside" and "inside" models and the final image tion lasts 3 minutes 35 seconds.
resolution was 768 by 460 pixels and 24 bit colour (8 bits
each for red, green and blue). Each image took typically 1

22
 THE DRESDEN FRAUENKIRCHE — REBUILDING THE PAST

Figure 3.9: Image of the "inside" pulpit and organ Figure 3.10: Image of the "inside' painted wooden dome
rendered using TDImage. rendered using TDImage.

3w3. Hardware is intended that future projects will use the power of the
PVS during the computer intensive rendering stage as well.
3.3.1. Computers
Two main computer environments were used for the project 3.3.2. Video
which, together with the Video and Slide making equip- A Supernova was connected to the same local area network
ment, were connected on the same local area network. The via a workstation. Each frame in the final animation se-
data files containing the geometric model, the photo realis- quence was read from the PVS and converted from compu-
tic model, the scanned and rendered images were shared ter graphics format to a format acceptable to the Supernova.
using the network. This ensured that all of the data were The Supernova performed the digital to analogue conver-
accessible to all pieces of equipment and only one copy of sion and sent each frame to a Betacam-SP video recorder.
the data was necessary. It took approximately 35 seconds to record each frame onto
The first environment consisted initially of four, and video tape and nearly 30 hours were required to record the
subsequently of eight, IBM RISC System/6000 workstations complete animation sequence. Some additional fades and
(both model 530H and 560) running the AIX operating sys- the musical accompaniment were completed in a television
tem. For each major stage an appropriate number of editing suite.
workstadons were employed, the number depending on the
people skills available or the computer resources required. 3.3.3. Slides
The allocation was approximately as follows:
Selected frames from the animation were chosen to be re-
• Geometric model — one machine for one week
rendered for photographic slides. All of the computer graph-
• Photo realistic model — 2D — one machine for one
ics figures in this paper were printed from 4 by 5 inch
week
negatives which were produced using a QCR-Z digitising
• Photo realistic model — 3D — three machines for camera; the image resolution was 4096 by 2730 pixels and
six weeks
24 bit colour (8 bits each for red, green and blue). In addi-
• Rendering — eight machines for four weeks
tion several 35mm stereo pairs were produced using an PCR
• Computer Post-production Compositing — one ma-
digitising camera; the image resolution was 2048 by 1365
chine for one week
pixels and 24 bit colour. Both the QCR-Z and the PCR
In all a total of nearly 9000 machine hours were expended
were connected to the same local area network via a worksta-
over an elapsed time of twelve weeks.
tion. [Ed. note: colour figures were converted to B/W via a
The second environment comprised an IBM POWER Visu- process camera for this paper.]
alization System (PVS). This is a parallel machine with 32
processors, 1 Gigabyte of shared memory, a 21 Gigabyte
3.4. Summary
disk subsystem and a 24-bit frame buffer with the ability to
display rendered images at up to 30 frames per second. The This project to reconstruct the Dresden Frauenkirche can
PVS was used for the following stages: be viewed as a typical application of commercial computer
• Animation Preview graphics software and hardware to a specific project. By
• Computer Post-production Fades its very nature all the software, though rich in function,
• Storage of the final animation sequence had limitations and the pressures of budget and deadlines
The PVS dramatically reduced the time needed for these necessitated compromises at all stages of the project.
stages to a few days instead of the several weeks that would Thus, for example, it was always possible to improve
have been required using single processor workstations. It some aspect of the camera trajectory or of some particular

23
B.COLLINS ET AL.

detail of the geometric model or of the texture mapping. It IBM POWER Visualization System (model 3) which is
was always possible to see "flaws" in a particular image or a product of IBM Corporation
animation sequence and to be tempted to continually strive SUPERNOVA (model 24.2) which is a product of
for "perfection". Thus, if images with "buggy" pixels were Spaceward Ltd
generated, it was felt to be acceptable to use a palette editor BETACAM-SP (model BVW-75P) which is a product
for 5 minutes to "correct" the offending pixels to circum- of Sony Corporation
vent the software problem rather than re-render the image, QCR-Z which is a product of Agfa Division of Miles
which could easily have taken one hour. The recent texts Inc.
(Vince, 1992, Watt and Watt, 1992) and the large body of PCR which is a product of Agfa Division of Miles
literature, see (Magnenat-Thalman and Thalman, 1992), Inc.
on computer animation though excellent on the theoretical Figs. 3.2 and 3.3 are reproduced by kind permission of the
aspects rarely discuss the compromises that are always nec- Deutsche Fotothek Dresden. Figures 3.1 and 3.4 are repro-
essary in a typical project. duced by kind permission of the Sachsen Landesamt für
Nevertheless, we believe that this computer re- Denkmalpflege.
construction recreates the original glory of the Dresden
Frauenkirche; a "virtual reality". This animation will also Bibliography
serve as an excellent preview of the long-planned recon- BuRRiDGE, J. M. ET AL. 1989. "The WINSOM solid modeller and
struction, which will soon start in "reaUty". its application to data visualization", IBM Systems Journal
28(4): 548-568.
3.5. Postscript — How to preserve a REILLY, P.1989. "Data Visualization in Archaeology", IBM Sys-
computer reconstruction ? tems Journal 28(4): 569-579.
MAGNENAT-THALMAN, N. & D. THALMAN 1992. "Six-hundred in-
After the completion of this animation we discovered that dexed references on computer animation", Journal of Visu-
we had created a new problem, which is "How can a com- alization and Computer Animation 3(3): 147-174.
puter reconstruction, which is a new archaeological docu-
VINCE, J. 1992. 3D Computer Animation, Addison-Wesley.
ment in its own right be itself conserved for future
WATT, A. & M. WATT 1992. Advanced Animation and Rendering
generations?". Computer hardware and software evolve and
Techniques, Addison-Wesley.
change at such a rate that it will most probably be impossi-
ble to render these images on computer systems even in the WooDWARK, J. R. 1991. "Reconstructing History with Computer
near future. Graphics", IEEE Computer Graphics and Applications 11(1).'
A possible solution would be to preserve the computer 18-20.
system which was used for each animation that is produced,
but that would ultimately become too expensive to main- Brian Collins and Dave Williams
tain in working order. The only practical solution at present IBM UK Scientific Centre,
would be to render these images at the maximum resolu- Athelstan House, St Clement Street,
tion possible and record them onto a medium such as 35mm GB-S023 9DR Winchester
cine-film or CD-ROM which has a reasonably long shelf
life. We hope that this will be possible with this project. collinsb @ winvmd.vnet.ibm.com
davew @ winvmd.vnet.ibm.com
Acknowledgements
More details of all of the computer graphics techniques can Robert Haak, Martin Trux and Herbert Herz
be found in the documentation for the software and hard- IBM Deutschland Informationssysteme GmbH,
ware products which were used in this project. In particu- Pascalstr. 100,
lar: D-7000 Stuttgart 80
CATIA (version 3.2) which is a product of Dassault
Luc Genevriez, Pascal Nicot, Pierrick Brault and Xavier
Systèmes
NEFERTrri (version 2.4) which is a product of Little Coyere
Big One ARC (Audiovisuel Realisation Conseil),
TDIMAGE (version 3.0) which is a product of
24 rue de Pre St Gervais,
Thomson Digital Interactive F-93500 Pantin
DATA EXPLORER (version 1.2) which is a product of
Burkhard Krause, Jens Kluckow and Armin Paffenholz
IBM Corporation
AIX (version 3.2) which is a product of IBM Corpo- TransCAT Nord GmbH,
ration Otto-Hahn-Str. 20,
IBM RISC System/6000 (models 530H and 560) D-4600 Dortmund 50
which is a product of IBM Corporation

24