Dating Wall Constructions from Brick Sizes in the Flooded Basement of a WW-II Submarine Bunker for the Digitization of Cultural Heritage

Abstract

Results from the exploration of the flooded basement of the U-Boot Bunker Valentin are presented. The work is done in the context of the digitization of the bunker, which was constructed during World-War II using substantial amounts of forced labor and which is now a memorial. The presented research deals with walls in the basement for which it is of interest when they were constructed. Using a simple but efficient structured light system, the sizes of the bricks were measured. In combination with the changes in German standards for bricks after WW-II, this gives important clues about the time of construction of the investigated walls.

Full text

A. Birk, F. Buda, and T. Hansen
Dating Wall Constructions from Brick Sizes in the Flooded Basement of a WW-II Submarine Bunker for the Digitization
of Cultural Heritage
MTS/IEEE OCEANS, Limerick, 2023
@inproceedings{Valentin-Basement-BrickSizes-Oceans23,
author = {Birk, Andreas and Buda, Frederike and Hansen, Tim},
title = {Dating Wall Constructions from Brick Sizes in the Flooded Basement
of a WW-II Submarine Bunker for the Digitization of Cultural Heritage},
booktitle = {MTS/IEEE OCEANS},
year = {2023},
type = {Conference Proceedings}
}
Preprint. For the final version, please see the reference above.

Dating Wall Constructions from Brick Sizes in the
Flooded Basement of a WW-II Submarine Bunker
for the Digitization of Cultural Heritage
Andreas Birk, Frederike Buda and Tim Hansen
Constructor University Bremen
Robotics, Computer Science and Electrical Engineering
thansen@constructor.university, fbuda@constructor.university, abirk@constructor.university
Abstract—Results from the exploration of the flooded basement
of the U-Boot Bunker Valentin are presented. The work is
done in the context of the digitization of the bunker, which
was constructed during World-War II using substantial amounts
of forced labor and which is now a memorial. The presented
research deals with walls in the basement for which it is of
interest when they were constructed. Using a simple but efficient
structured light system, the sizes of the bricks were measured.
In combination with the changes in German standards for
bricks after WW-II, this gives important clues about the time
of construction of the investigated walls.
Index Terms—Cultural heritage, structured light, laser scan-
ner, measurement, underwater exploration, digital humanities,
digital history
I. INTRODUCTION
The research presented here is conducted in the context of
the digitization of the World-War-II submarine bunker Valentin
(Fig. 1) with air-, ground-, and underwater robots [1]. The
outside and the interior are 3D-modeled with photogramme-
try, respectively Lidar data using a spectral method [2] for
registration [3]. While according technologies are meanwhile
standard tools for the study of cultural heritage in general
[4]–[6] , their application is more challenging for according
underwater applications [7].
The bunker is nowadays a memorial, for which the digitiza-
tion is supported by the German Federal Ministry of Education
and Research within the eHeritage program. Its construction
took place from 1943 to 1945 with massive use of forced
laborers. It was the largest armament project of the German
navy and it was supposed to function as a submarine shipyard
for them. The plan was to produce Type-XXI submarines, but
this did not happen because the bunker could not be completed
before the end of World War II. Up to 8,000 forced laborers
worked on the bunker construction site every day, and many
of them lost their lives [1].
The bunker is a very large building. It has a length of 426
meters, a width of 67 meters (east) to 97 meters (west) and
a height of 30-33 meters. With its floor area of 35,375 m2, it
is the largest free-standing bunker in Germany and the second
largest in Europe after the submarine bunker in Brest, built in
occupied France. The interior, secured with 7 m thick floor
concrete and 4.5 m thick exterior walls, contains 520,000 m3.
Fig. 1. An aerial view on the bunker Valentin (top), which is modelled in
3D (bottom) using photogrammetry and 3D registration of Lidar data.
II. MOT IVATIO N AN D APP ROAC H
Major parts of the bunker are underwater, e.g., a large basin
that was planned for the release of submarines into the Weser
river from which they could move to the sea [8]. But also the
basement of the bunker is flooded and it is unclear to which
extent this is intentional or not and when this happened. Also,
there are discrepancies between the construction plans and
actual implementations, which are of interest for historians.
The exploration of the basement is impossible for human
divers as it is much too dangerous.
In the flooded basement are multiple small walls, which
differ significantly from the main concrete walls of the bunker
as they are constructed with bricks. These bricks can aid the
dating of their construction and hence of the time the basement
was not flooded. This is due to the fact that the standards

Fig. 2. The visibility is very poor in the flooded basement, i.e., it is Zero for significant amounts of time during the exploration with the robot (left). The
camera only provides useful information when being very close to structures, e.g., a wall fragment (right).
Fig. 3. Map of a large room in the flooded basement generated from over 20,000 consecutive measurements with a Ping360 sonar on a BlueROV.
for brick sizes have changed in Germany after World-War-II
(Tab.I).
For the exploration of the flooded basement where the walls
are located, a BlueROV21is used, which is equipped with a
low-cost scanning sonar in form of a Ping3602and a small
WaterLinked A50 Doppler Velocity Log (DVL)3to aid the
navigation under poor visibility conditions (Fig.2), especially
by creating maps (Fig.3) [9].
More important from the perspective of the research pre-
sented here, the robot is equipped with a self-sufficient laser-
system for measuring the brick-sizes.
The use of structured light is a very well-established tool
for underwater applications since several decades [10]–[13].
The aspect of interest presented here is the specific use-case,
1https://bluerobotics.com/store/rov/bluerov2/
2https://bluerobotics.com/store/sensors-sonars-cameras/sonar/ping360-
sonar-r1-rp/
3https://www.waterlinked.com/dvl/dvl-a50
for which the system is adapted as a quite simple, but highly
efficient self-contained solution.
III. SEN SO R AN D MET HO DS
The laser-system consists of 4 red (650 nm) point-lasers
with 100 mW each. The lasers are mounted in a 3D-printed
holder that is fitted into a very small and light-weight bottle
with a flat-pane acrylic window (Fig. 5, left). The bottle can
be mounted such that the 4 red lasers mark the corners of a
uni-distant cross or of a square. As the lasers are perpendicular
to the flat port, there is no need to take the specific refraction
effects into account [14].
The bottle has an inner diameter of 50 mm and a length of
100 mm. In addition to the lasers, a suited switching power
supply in form of a XP Power VR10S3V3 and a high-capacity
Li-Thionylchlorid battery are fitted into the bottle. The battery
is rated at 3.6 V with 2.1 Ah, i.e., it provides 7.56 Wh or 18.9
hours of continuous operation of the lasers. The laser-bottle

Fig. 4. An overview of the methods used for the processing.
hence does not need any power from the ROV and can hence
be easily attached to the vehicle without any cabling.
The bottle is mounted with a 3D-printed holder on the
payload skid of the BlueROV2 (Fig. 5, right) in an exactly
known location, such that the laser-beams are perpendicular to
the image plane. Given the distances of 30 mm in the diagonal
between the lasers, the plane-homography of a wall in front
of the camera can be computed.
Concretely, the images are post-processed using standard
image processing methods [15], [16] implemented in the well-
known OpenCV library [17] using the Python interface. Fig. 4
gives an overview of the methods used.
Most importantly, the lasers can be used to establish the
basis for metric measurements within the monocular camera
images: First, the known camera parameters are used to (1)
rectify the image. This is followed by a (2) Gaussian blur with
a small 5×5 kernel. The image is then converted into grey-
scale by (3) selecting the V-channel after a color-conversion
from RGB to HSV. A simple (4) thresholding operation turns
the image into a binary representation as the lasers are always
by far the brightest regions. A (5) contour detection on the
laser blops is then the basis to do a (6) centroid detection.
Finally, there is a (7) sanity check that ensures that there
are exactly 4 centroids that establish reasonable geometric
constraints.
Fig. 6 illustrates the detection of the laser centroids with a
typical example. Given the geometric arrangement of the lasers
and the camera, the (8) plane homography of the wall can
be computed. This is done using the Python Linear Algebra
package.
Furthermore, it provides the basis to measure the bricks.
So, the next step is to (9) calculate metric distances in the
pixel space, i.e., to use the known distances between the lasers
and the homography to calculate the millimeters per pixel
width, respectively height. To aid the visibility of the bricks in
the image, a (10) contrast enhancement of the original image
under the homography can be applied (Fig. 6, right).
At that stage, it is actually quite fast and accurate to
simply do a (11-A) manual measurement for each brick. The
alternative is to use a (11-B) fully automated pipeline for which
we denote its sub-steps with (OX) as they are optional.
First, a (O1) texture-based segmentation is applied to the
original image transformed by the homography. A (O2) Lapla-
cian edge detection is then used to extract the contours
of the bricks. To get their boundaries as straight lines, a
(O3) vertical/horizontal Hough transform is performed, i.e.,
a Hough transform where only lines with (approximately)
horizontal or vertical orientation are considered. As a further
sanity check, a minimum number of votes in the Hough space
is required to be considered to be a solid basis for determining
a brick size through the intersections of the lines.
Fig.7 shows an example for a single segment, i.e., one
candidate brick. The brick segments (left) are in white, the
Laplacian edges are shown in green (center), and the four
vertical/horizontal lines that form the boundaries of the ex-
ample brick segment are shown in red. The fully automated
pipeline works reasonably well, but depending on adjusting
the parameters in the sanity checking, it either computes the
sizes accurately but for significantly less bricks than a human
can identify or there can be some outliers.
IV. RES ULT S
For the analysis, 937 images with 7,393 bricks in total were
processed. For the fully automated pipeline, the automated
sanity checks were strictly set to avoid any outliers. Hence,
while about 15 bricks are detectable on average per image for a
human, only about 8 bricks are automatically processed in the
fully automated pipeline. As an additional sanity check, 200
images or about 21% were processed by a human to validate
the automated processing.
As mentioned, the main question of interest is when the
brick-walls in the flooded basement of the bunker were con-
structed. Tab. I shows selected German standards for bricks.
After WW-II, the modern standard based on the Normalformat
(NF) was introduced that is still in use today. It can be
literally translated as ”regular size”, on which a few variations
are based, namely D¨
unnformat (DF),Zweifaches D¨
unnformat
(DF), and Langd¨
unnformat (DF) that differ especially in
height from the regular size.
The modern standard NF and its variants are used all across
Germany. Before and during WW-II, there was a nationwide
standard in form of the Reichsformat (RF), but regional
standards were also widely in use. In addition to the RF, a

Fig. 5. CAD drawing of the laser bottle (left) and its mounting on the BlueROV2 vehicle (right).
Fig. 6. An input image (left) and the centroids of the laser blobs indicated via the center of green circles (center). The centroids can be used to calculate the
homography of the plane of the wall. An additional contrast enhancement can be used to better highlight the bricks (right).
Fig. 7. An example of the determination of the brick size in a fully automated pipeline with (O1) texture-based segmentation (left), (O2) Laplacian edge
detection (center), and (O3) vertical/horizontal Hough transform (right).
TABLE I
AN OVE RVIE W OF SE LE CTE D GERMAN BRICK STANDARDS DURING AND AFTER WW-II.
Dimensions (mm) Aspect Ratios
L W H L/H W/H
historical
Elbformat 230 110 52 4.42 2.12
Friesenziegel 206 100 51 4.04 1.96
Hamburger Format 220 105 65 3.38 1.62
Oldenburger Format 220 105 52 4.23 2.02
Reichsformat (RF) 240 115 63 3.81 1.83
modern
D¨
unnformat (DF) 240 115 52 4.62 2.21
Zweifaches D¨
unnformat (2DF) 240 115 113 2.12 1.02
Langd¨
unnformat (LDF) 290 115 52 5.58 2.21
Normalformat (NF) 240 115 71 3.38 1.62

Fig. 8. The bricks show a high amount of variation. Type (A) and (C) are the dimensions of the historic Oldenburger Format in terms of length and height,
respectively width and height. The others fit in terms of height, but neither length nor width.
few other historical standards, especially ones with a link to
Northern Germany where Bremen is located, are shown in
Tab. I.
In addition to the absolute dimensions, the aspect ratio
of the length, respectively of the width to the height of the
bricks are also of interest. They provide an additional source
of information that is independent of the absolute metric
scale. They could hence already be used in combination with
a calibrated monocular camera. But the differences between
them are relatively small; hence a measurement error of a few
pixel would already lead to inconclusive results when using
only scale-free measurements.
TABLE II
THE NUMBER OF BRICKS THAT COMPLETELY CORRESPOND TO THE
OLD ENB URG ER FORMAT,EITHER WITH LENGTH AND HEIGHT OR WITH
WIDTH AND HEIGHT,PLUS T HE N UMB ER O F BRI CK S THAT ON LY MATCH
WITH THEIR HEIGHT (OTH ER) .
#bricks percentage
Oldenburger Format (length) 4550 61.5%
Oldenburger Format (width) 1456 19.7%
other 1387 18.8%
The analysis of the brick sizes shows (a) a high variance that
(b) is distributed around several modes that do not correspond
to the dimensions of the German brick standard after World-
War-II. This indicates that the bricks originated during the
time of the construction of the bunker, respectively that they
may have even been fabricated on the construction site of the
bunker itself.
More precisely, a significant number of bricks, namely
81.2%, corresponds to the Oldenburger Format. As of course
only two of the three dimensions are visible for each brick,
this total number is a combination of bricks where the height
is according to this standard plus either the length or width.
The related absolute numbers as well as the percentages are
shown in Tab. II. This is already a strong indication that the
walls were constructed during the time of WW-II, i.e., as part
of the construction of the bunker.
There is also a substantial number of bricks that is not a
complete match to the Oldenburger Format - or other likely
candidate standards. At least, their height tends to corresponds
to the Oldenburger Format, but their length or width does not.
Fig. 8 provides an illustrative example. There are different
possible explanations.
First, they may be from a different production, which
follows a standard that has not been taken into consideration.
This is very unlikely as all the bricks in the different walls look
very similar, especially with respect to their color. Bricks that
were baked at different locations tend to have clearly visible
differences caused by the composition of the raw material, its
handling, the baking process, etc.
Second, the bricks may simply be defective, e.g., broken
during or after the production process or transport. This
hypothesis is not completely unlikely. Especially, it is not fully
clear what the purpose of the walls was; they for example do
not appear on the known construction plans of the bunker. It
may be that the basement was used as an air-raid shelter and
the walls were constructed in this context. It is likely that no
prime building material but also defective bricks were used
for this purpose. But the edges of these bricks seem to be
very straight; there are no clearly visible signs of cracks or
breaking that one may expect in this case.
Third, the production itself may have been flawed, respec-
tively happening in a make-shift fashion. The construction
site of the bunker included multiple larger work-shops and
facilities. Possibly, the bricks were even produced on or near
the construction site.
V. CONCLUSIONS
A use case of underwater structured light in the context
of digitization of cultural heritage was presented. Concretely,
a simple but effective laser-system was used to determine
the sizes of bricks in wall-structures in the flooded basement
of the memorial U-Boot bunker Valentin. Given the changes
in standards for brick sizes in Germany after World-War
II, this information provided some clues about the time of
construction of the walls. The results show that a significant

numbers of bricks is in the historic ”Oldenburger Format”, i.e.,
a standard that predates the modern ”Normal Format” used
after WW-II. This suggests that the walls were build during
the construction of the bunker. Furthermore, there are multiple
bricks that correspond with the height of the ”Oldenburger
Format”, but not with its length or width. While a detailed
explanation of this observation still needs some investigation
on the historical side, there is a relatively high likelihood that
it is linked to the circumstances in which the bunker was built
with many makeshift arrangements.
ACK NOW LE DG ME NT S
The presented work was conducted within the project ”3D
Mapping of the U-Boot Bunker Valentin memorial by Air-
, Ground-, and Underwater-Robots (Valentin3D)” supported
by the German Federal Ministry of Education and Research
(BMBF) in the framework of the promotion of research and
development projects for the digitization of cultural objects
(eHeritage).
REFERENCES
[1] A. Birk, F. Buda, H. B ¨
ulow, A. G. Chavez, C. A. M¨
uller, and J. Timpe,
Digitizing a Gigantic Nazi Construction: 3D-Mapping of Bunker
Valentin in Bremen. Berlin, Boston: De Gruyter, 2022, pp. 133–168.
[Online]. Available: https://doi.org/10.1515/9783110714692-006
[2] H. B¨
ulow and A. Birk, “Scale-free registrations in 3d: 7 degrees of
freedom with fourier-mellin-soft transforms,” International Journal of
Computer Vision (IJCV), vol. 126, no. 7, pp. 731–750, 2018.
[3] H. B¨
ulow, C. A. M¨
uller, F. Buda, A. G. Chavez, and A. Birk, “A divide
and conquer method for 3d registration of inhomogeneous, partially
overlapping scans with fourier mellin soft (fms),” in IEEE International
Conference on Robotics and Automation (ICRA), 2020, Conference
Proceedings.
[4] V. Poulopoulos and M. Wallace, “Digital technologies and the role of
data in cultural heritage: The past, the present, and the future,” Big
Data and Cognitive Computing, vol. 6, no. 3, p. 73, 2022. [Online].
Available: https://www.mdpi.com/2504-2289/6/3/73
[5] M. K. Bekele, R. Pierdicca, E. Frontoni, E. S. Malinverni, and
J. Gain, “A survey of augmented, virtual, and mixed reality for cultural
heritage,” J. Comput. Cult. Herit., vol. 11, no. 2, p. Article 7, 2018.
[Online]. Available: https://doi.org/10.1145/3145534
[6] F. Galeazzi, “Towards the definition of best 3d prac-
tices in archaeology: Assessing 3d documentation techniques
for intra-site data recording,” Journal of Cultural Her-
itage, vol. 17, pp. 159–169, 2016. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S129620741500120X
[7] F. Menna, P. Agrafiotis, and A. Georgopoulos, “State
of the art and applications in archaeological underwater
3d recording and mapping,” Journal of Cultural Her-
itage, vol. 33, pp. 231–248, 2018. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S1296207417308555
[8] T. Hansen and A. Birk, “Using registration with fourier-soft in 2d (fs2d)
for robust scan matching of sonar range data,” in IEEE International
Conference on Robotics and Automation (ICRA), 2023, Conference
Proceedings.
[9] T. Hansen, F. Buda, and A. Birk, “Underwater exploration with sonar
of the flooded basement of a ww-ii submarine bunker in the context
of digitization of cultural heritage,” in MTS/IEEE OCEANS, 2023,
Conference Proceedings, pp. 1–5.
[10] M. Castillon, A. Palomer, J. Forest, and P. Ridao, “State of the art of
underwater active optical 3d scanners,” Sensors, vol. 19, no. 23, pp.
1–35, 2019. [Online]. Available: https://doi.org/10.3390/s19235161
[11] S. Tetlow and J. Spours, “Three-dimensional measurement of underwater
work sites using structured laser light,” Measurement Science and
Technology, vol. 10, no. 12, pp. 1162–1167, 1999. [Online]. Available:
http://dx.doi.org/10.1088/0957-0233/10/12/307
[12] E. Trucco, R. B. Fisher, A. W. Fitzgibbon, and D. K.
Naidu, “Calibration, data consistency and model acquisition with
laser stripers,” International Journal of Computer Integrated
Manufacturing, vol. 11, no. 4, pp. 293–310, 1998. [Online].
Available: https://doi.org/10.1080/095119298130642
[13] F. M. Caimi, D. M. Kocak, and V. L. Asper, “Developments in laser-
line scanned undersea surface mapping and image analysis systems
for scientific applications,” in MTS/IEEE OCEANS, 1996, Conference
Proceedings, pp. 75–81 suppl.
[14] T. Luczynski, M. Pfingsthorn, and A. Birk, “The pinax-model for
accurate and efficient refraction correction of underwater cameras in
flat-pane housings,” Ocean Engineering, vol. 133, pp. 9–22, 2017.
[15] R. Szeliski, Computer Vision: Algorithms and Applications, ser. Texts
in Computer Science, 2022.
[16] E. R. Davies, Computer Vision: Principles, Algorithms, Applications,
Learning. Academic Press, 2017.
[17] G. B. A. Kaehler, Learning OpenCV. O’Reilly Media, 2008.