ArticlePDF Available

Photogrammetric Reconstruction of the Great Buddha of Bamiyan, Afghanistan

Authors:

Abstract and Figures

In the valley of Bamiyan, Afkhanistan, approximately 1700 years ago, two large standing Buddha statues were carved out of the sedimentary rock of the region. They were 53 and 38 in high and the larger one-figured as the tallest representation of a standing Buddha in the world. In March 2001 the Taleban government militia demolished the colossal statues. After the destruction a group from ETH Zurich completed the computer reconstruction of the Great Buddha, which can serve as the basis for a physical reconstruction. This paper reports the results of the image-based 3D reconstruction of the statue, performed on three different data-sets in parallel and using different photogrammetric techniques and algorithms.
Content may be subject to copyright.
PHOTOGRAMMETRIC RECONSTRUCTION OF THE
GREAT BUDDHA OF BAMIYAN, AFGHANISTAN
Armin Gru
¨n(agruen@geod.baug.ethz.ch)
Fabio Remondino (fabio@geod.baug.ethz.ch)
Li Zhang (zhangl@geod.baug.ethz.ch)
Swiss Federal Institute of Technology (ETH), Zurich
Abstract
In the valley of Bamiyan, Afghanistan, approximately 1700 years ago,
two large standing Buddha statues were carved out of the sedimentary rock of
the region. They were 53 and 38 m high and the larger one figured as the
tallest representation of a standing Buddha in the world. In March 2001 the
Taleban government militia demolished the colossal statues. After the destruc-
tion a group from ETH Zu¨rich completed the computer reconstruction of the
Great Buddha, which can serve as the basis for a physical reconstruction.
This paper reports the results of the image-based 3D reconstruction of the
statue, performed on three different data-sets in parallel and using different
photogrammetric techniques and algorithms.
Keywords: 3D reconstruction, Bamiyan Buddhas, image matching,
modelling, orientation, visualisation
The Buddha, the Godhead, resides quite as comfortably in the circuits of a
digital computer (or the gears of a cycle transmission) as he does at the top
of a mountain or in the petals of a flower. (Robert Pirsig, Zen and the Art of
Motorcycle Maintenance, W. Morrow & Company London, 1974, 432 pages)
Introduction
The region of Bamiyan, approximately 200 km north-west of Kabul, Afghanistan,
was one of the major Buddhist centres from the 2nd century ad up to the time when
Islam entered the area in the 8th century. For centuries, Bamiyan lay at the heart of the
famous Silk Road, offering rest to caravans carrying goods across the area between
China and Western empires. Strategically situated in a central location for travellers
from north to south and east to west, the village of Bamiyan was a common meeting
place for many ancient cultures.
In the Bamiyan valley (Fig. 1), at an altitude of 2500 m, three big statues of
Buddha and a great number of caves were carved out from the sedimentary rock of the
region (Figs. 2 and 3). There were two big standing Buddha statues, which stood about
900 m apart, while in the centre there was a smaller image of a seated Buddha (Fig. 5).
The Photogrammetric Record 19(107): 177–199 (September 2004)
2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd.
Blackwell Publishing Ltd. 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street Malden, MA 02148, USA.
The Buddhas were built under the control of the Kushan dynasty, which ruled
between the late 1st century and early 3rd century ad over a kingdom incorporating
Northern India, certain regions of Central Asia and areas corresponding to present-day
Afghanistan and Pakistan. The Emperor Kanishka ordered the construction of the
statues and some descendants of Greek artists, who went to Afghanistan with
Alexander the Great, started the construction that lasted probably until the 3rd or 4th
century ad. The Kushan dynasty produced the distinctive Gandhara art. Gandhara
developed an artistic style blending Greco-Roman influences and Indian Buddhism.
The Muse´e National des Arts Asiatiques-Guimet holds many spectacular
pieces of art from that period of great artistic refinement (Muse´e National
des Arts Asiatiques-Guimet, 2004).
These objects of religious sculpture, including the Buddhas of Bamiyan, belong to the
Indian Mathura school. J. Hackin even argues that:
Nothing could be more natural than that the artists of Mathura were inspired
by the statues at Bamiyan justly famous at that time throughout the Buddhist
world (Hackin, 1928: page 109).
The larger statue of Bamiyan was 53 m high while the smaller one measured 38 m
(Fig. 3). The Great Buddha should represent Vairocana, the ‘‘Light Shining throughout
Fig. 1. The Bamiyan valley, as seen by Ikonos
(courtesy of Space Imaging, Inc., Denver, CO (Grant Street)).
Gru
¨net al. Photogrammetric reconstruction of the Great Buddha of Bamiyan, Afghanistan
178 2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd.
the Universe’’ Buddha, while the small one should represent Shakyamuni (AIIS, 2004;
IRAO, 2004). They were cut from the sandstone cliffs and were covered with a mud
and straw mixture to model fine details such as the expression of the face, the hands
and the folds of the robe. An account of the building technique of the colossal statue is
given in Knobloch (2002: page 93):
The Bamiyan Buddha was created by cutting a high-relief figure into the face
of the soft conglomerate cliff. It is possible that the niche was carved out
first, using scaffolding slotted into holes cut into the cliff, before the
ambulatory galleries were carved; the scaffolding later being replaced by a
Fig. 3. The Great Buddha (left and centre) and the smaller statue (right).
Fig. 2. The empty Buddha caves as observed by Ikonos (courtesy of Space Imaging, Inc.). Left: Great
Buddha. Right: Small Buddha.
The Photogrammetric Record
2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd. 179
series of permanent wooden ladders, landings and facades. The torso was
roughly shaped and detailing of the folds of the gown was built out by cutting
lines of shallow holes for wooden pegs on which were hung ropes coated
with thick stucco.
The lower parts of the arms were constructed on wooden armatures while it is
generally assumed that the upper parts of the faces were made as wooden masks. The
two giants were painted in gold and other colours and they were decorated with
dazzling ornaments. They are considered the first series of colossal cult images in
Buddhist art and may even be among the first representations of Buddha himself in
human form, replacing the older symbolic portrayals of the Indian bas-reliefs. The
niches of the statues, as well as most of the caves of the cliff, were also decorated with
colourful frescos.
First written reports about the Buddha statues come from the Chinese travellers
Fah-Sien (around ad 400) and Hsuan-Tsang (ad 630). Especially, Hsuan-Tsang
describes in great detail the site where 5000 monks were active in monasteries and
caves, carved into the large rock face, containing also two standing Buddhas. He also
reports of the Great Buddha being painted in vivid colours and having a wooden face
mask painted in gold.
The first damage came about with the arrival of Islam in the 8th century.
Reportedly, Genghis Khan destroyed the town of Bamiyan in 1221, but did not do any
harm to the monks and the Buddhas. Major destruction through the firing of cannon
balls at the statues is reported under the Moguls Shah Aurangzeb and Nadir Shah in the
17th and 18th centuries, respectively.
The 19th century saw an influx of amateur archaeologists, who more specifically
were medical doctors, military personnel, government agents and the occasional
traveller. Alexander Burnes, who visited Bamiyan in 1832, is considered the modern
discoverer of the Buddhas, although he was not the first eyewitness of modern times.
Fig. 4 shows a drawing by Burnes in which the Buddhas have a fantastic appearance
and only little resemblance with reality.
Fig. 4. Alexander Burnes’ drawing of the Buddhas of Bamiyan.
Gru
¨net al. Photogrammetric reconstruction of the Great Buddha of Bamiyan, Afghanistan
180 2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd.
In 1833 the military deserter, secret service agent and treasure hunter James Lewis
(who called himself ‘‘Charles Masson’’) visited the site. It is said that he left the
following graffito (which however could not be found by the present authors): ‘‘If any
fool this high samooch explore know Charles Masson has been here before.’
It took another century until the first serious excavations and investigations were
conducted. For the French agency DAFA (Delegation Arche´ologique Franc¸aise en
Afghanistan) renowned experts like A. Godard, J. Carl and J. and R. Hackin worked
between 1920 and 1930 in Bamiyan. R. Hackin published the first guidebook on
Bamiyan in 1934. He noticed inexperienced craftsmanship on the Small Buddha as
well as some cumbersome and primitive features. Since the Great Buddha is more
sensibly proportioned, it was assumed that it was built after the Small Buddha. But
there is no clear evidence for this.
Robert Byron, the English travel writer who visited Bamiyan in 1933/34
obviously disliked the Buddhas very much. He wrote in his The Road to Oxiana:
I should not stay long at Bamian. Its art is unfresh Neither [Buddha] has
an artistic value. But one could beat that: it is their negation of sense, the
lack of any pride in their monstrous flaccid bulk that sickens.
But then he went on to marvel about the landscape:
The colours of this extraordinary valley with its cliff of rhubarb red, its
indigo peaks roofed in glittering snow and its new-sprung corn of harsh
electric green, shone doubly brilliant in the clear mountain air And there
suddenly, like an enormous wasps’ nest, hung the myriad caves of the
Buddhist monks, clustered about the two giant Buddhas. (Byron, 1981)
Violent events since 1979 have damaged both society and infrastructure in
Afghanistan and have greatly reduced the number of visitors to Bamiyan. The long period
of conflict culminated in an edict by the Taleban government to destroy non-Islamic
images in the country. Despite major international efforts, notably by ICOMOS (2001)
and UNESCO, to persuade the government to leave such works of world cultural heritage
unharmed, or to accept the building of walls which would leave them merely hidden, the
Bamiyan statues were demolished by Taleban forces in March 2001 (Figs. 5 and 6).
The quality of the Buddha statues can no longer be argued about, because they are
gone. But the valley of Bamiyan and its surroundings, which we visited on a
photogrammetric field campaign in August 2003, is one of the most beautiful sites and
spectacular views of this world.
After the destruction, an intensive discussion started at an international level
concerning the need for a physical reconstruction of the statues. The matter is not yet
resolved, although many recent signs point towards a reconstruction (UNESCO, 2004)
(Fig. 6).
ETH Zu¨rich has volunteered to perform the computer reconstruction, which can
serve as a basis for the physical reconstruction. In fact, using a computer model, a
statue at 1/10 of the original size will first be built and displayed in the Afghanistan
Museum in Bubendorf, Switzerland. But the most recent developments actually call for
the placement of this model into the National Museum of Kabul, Afghanistan.
Originally, ETH Zu¨ rich interest in the reconstruction of the Great Buddha was
purely scientific. It was planned to investigate to what extent such an object could be
The Photogrammetric Record
2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd. 181
reconstructed fully automatically, simply by using amateur images taken from the
Internet. The main scientific challenge here lies in the fact that no typical
photogrammetric information (such as interior and exterior orientation parameters)
about these images is available and that existing automated image analysis techniques
will most probably fail under the given circumstances, as described later. After learning
about the efforts to actually rebuild the Great Buddha it was decided to get involved in
that project beyond a purely scientific approach in order to contribute as much as
possible to the success of the work through the use of ETH Zu¨rich’s technology.
Finally, three sets of images were available for the computer reconstruction: Internet
images, amateur images from a visiting tourist and metric images. The resulting
computer models of the Buddha varied in accordance with the images used in each
case. The results extracted from the Internet and tourist images served only for
scientific purposes and to test the newly developed matching algorithm. The physical
reconstruction will be based on a 3D computer model derived from three metric
images. These images were acquired in Bamiyan in 1970 by Professor Kostka,
Technical University of Graz (Kostka, 1974). They form the basis for a very precise,
reliable and detailed reconstruction with an accuracy of 1 to 2 cm in relative position
and with an object resolution of about 5 cm. Manual image measurements had to be
applied in order to achieve these values.
Fig. 6. Left: The explosion of the big statue (courtesy of CNN). Centre: The empty cave after the
destruction. Right: The stones of the big statue recovered and protected with UNESCO bags.
Fig. 5. A panorama of the cliff of Bamiyan valley after the destruction with three images showing the
Buddha statues prior to demolition.
Gru
¨net al. Photogrammetric reconstruction of the Great Buddha of Bamiyan, Afghanistan
182 2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd.
In this paper are presented the results of the computer reconstruction of the Great
Buddha obtained from different data-sets of images and using different photogram-
metric algorithms. For each set, the reconstruction process consists of photo-
triangulation (calibration, orientation and bundle adjustment), image coordinate
measurement (automatic matching or a manual procedure), point cloud and surface
generation, texture mapping and visualisation. From Kostka (1974) a contour plot of
the larger statue is also available (20 cm isolines, scale 1:100) and some control points
were derived from it for use in the photo-triangulation process.
Available Images of the Great Buddha
The work is based on three different types of imagery, used in parallel:
(1) A set of images acquired from the Internet (‘‘Internet images’’).
(2) A set of tourist photographs acquired in the valley of Bamiyan between 1965
and 1969 (‘‘Tourist images’’).
(3) Three metric images acquired in 1970 by Professor Kostka, Technical Uni-
versity of Graz (Kostka, 1974).
Of the 15 images found on the Internet, four were selected for processing (Fig. 7):
two in front of the big Buddha, one from the left side and one from the right side of the
statue. All the others were less suitable for photogrammetric processing because of
very low image quality, occlusions or small image scale.
Fig. 7. The Internet images used for the 3D reconstruction.
Fig. 9. The four tourist images (1840 ·1232 pixels) used for the computer reconstruction.
The Photogrammetric Record
2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd. 183
The main problems with these images are their differences in size and scale, the
unknown pixel size and camera constants and most of all the different times of
acquisition; therefore some parts visible in one image are missing in others (Fig. 8).
Also the illumination conditions (shadows) are very different and this can create
problems with automatic matching procedures.
The tourist images were provided by Harald Baumgartner, who visited the valley
of Bamiyan at the end of the 1960s. They were slides, acquired with an Asahi Pentax
camera and then scanned at a resolution of 50 lm. Out of 12 available images, four
were selected for processing (Fig. 9). Also in this case, the illumination conditions
were not constant in all the images and the orientation parameters of the cameras were
not known.
The metric images were acquired in August 1970 with a TAF camera
(Finsterwalder, 1896; Finsterwalder and Hofmann, 1968). The TAF (Terrestrische
Ausru¨stung Finsterwalder) is a phototheodolite camera (Fig. 16, left) that acquires
photos on 13 ·18 cm glass plates. Two fiducial marks are present on the longer sides
of the photos while a pointer defines the image horizon with an index that moves
vertically.
The original photos were scanned by Vexcel Imaging Inc. Graz, Austria with the
ULTRA SCAN 5000 at a resolution of 10 lm. The digitised images each resulted in
16 930 ·12 700 pixels (Fig. 10).
Image Measurement with Least Squares Matching Algorithm
A multi-photo geometrically constrained (MPGC) least squares matching
software package, developed at the Institute of Geodesy and Photogrammetry at
ETH Zu¨rich, was applied to all three image data-sets (Gru¨ n et al., 2001, 2003).
(a) (b) (c)
Fig. 10. The three metric images acquired in Bamiyan in 1970 by Professor Kostka.
Fig. 8. Changed details between the images (red circles) and different illumination conditions (right).
Gru
¨net al. Photogrammetric reconstruction of the Great Buddha of Bamiyan, Afghanistan
184 2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd.
The automatic surface reconstruction (Fig. 11) works in fully automated mode
according to the following procedure:
(1) Selection of one image as the master image.
(2) Extraction of a very dense pattern of feature points in the master image using
the Fo¨rstner operator. At this stage, the master image was subdivided into
7·7 pixel image patches and within each patch only the single feature point
which gained the highest interest value was selected. In the implementation,
the threshold for the Fo¨rstner parameter roundness was set to 0Æ65 and the
grey value variance of the image window was not allowed to drop below 5Æ0.
(3) For each feature point, using the epipolar geometry determined by photo-
triangulation, the approximate matches were obtained for the following
MPGC matching procedure by standard cross-correlation. The threshold of
the normalised correlation coefficient is usually set to 0Æ7. The position and
size of the search areas are determined by using the already known
approximate surface model. In order to get this approximate surface, image
pyramids and a matching strategy based on region growing, which takes the
already measured control points as seed points, are used.
(4) In the final step MPGC matching is applied for fine matching, including patch
reshaping parameters (Gru¨n and Baltsavias, 1988; Baltsavias, 1991). MPGC
exploits a priori known geometric orientation information to constrain the
solution and allows for the simultaneous use of more than two images. The
Fig. 11. Data flow diagram of the matching procedure.
The Photogrammetric Record
2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd. 185
algorithm can use blunder and occlusion detection and quality control by self-
diagnosis.
MPGC is a non-linear algorithm. Its linearisation requires the procedure to be
iterated. Considering the need to speed up the convergence rate, the geometric
constraint is usually applied in a fully constrained mode in the first and second
iterations. Then the weights of the geometric constraints are decreased in order to
consider the errors in orientation data. To reduce the oscillations and divergence
problems the reshaping parameters can be constrained according to the image signal
content in the image patches and based on the analysis of the covariance matrix of least
squares matching.
With the MPGC approach, sub-pixel accuracy matching results and 3D object
coordinates can be obtained simultaneously (Fig. 12) and also, through covariance
matrix computations, a good basis for quality control.
The matching procedure previously described has also a modified version that is
used when the images have large scale and/or rotation differences between each other.
In these cases the cross-correlation technique used in the first step to extract the
approximations for the MPGC matching normally encounters difficulties. In order to
solve these restrictions rectified images are used. The rectified images are generated
using a vertical plane fitted to the control and tie points (Fig. 13, left). Then each pixel
of the original image is projected onto the vertical plane by using the known
orientation parameters. The relationships between the original and rectified images can
be calculated using the collinearity equations:
Xp¼ðZcZ0Þr11xpþr21 ypr31f
r13xpþr23 ypr33fþX0
Yp¼ðZcZ0Þr12xpþr22 ypr32f
r13xpþr23 ypr33fþY0
ð1Þ
Fig. 12. Epipolar geometry between the metric images to get the correct matching (upper row). MPGC
matching results (patch resampling) and computed 3D object coordinates (lower row).
Gru
¨net al. Photogrammetric reconstruction of the Great Buddha of Bamiyan, Afghanistan
186 2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd.
xp¼fr11ðXpX0Þþr12 ðYpY0Þþr13 ðZcZ0Þ
r31ðXpX0Þþr32 ðYpY0Þþr33ðZcZ0Þ
yp¼fr21ðXpX0Þþr22 ðYpY0Þþr23 ðZcZ0Þ
r31ðXpX0Þþr32 ðYpY0Þþr33ðZcZ0Þ
ð2Þ
where Z
c
is the constant value of the vertical plane; (X
0
,Y
0
,Z
0
) is the position of the
perspective centre; (r
11
,r
12
,,r
33
) is the rotation matrix calculated from the attitude
values of images; fis the camera constant; and (X
p
,Y
p
) and (x
p
,y
p
) are the pixel
coordinates of the rectified and original images, respectively. After generation of the
rectified images, feature points are extracted in one of them (master image) and their
conjugate points in other rectified images are computed by standard cross-correlation
Fig. 13. Definition of the vertical plane with constant Z
c
, used for image rectification (left). Calculation
of the initial reshaping parameters of MPGC. S,S¢are the perspective centres of the template and search
image, respectively (right).
Fig. 14. Recovered camera poses and object coordinates of the Internet images (left). Point cloud (centre)
and 3D model of the Buddha displayed in textured mode (right images).
The Photogrammetric Record
2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd. 187
and a matching strategy based on region growing. The already measured control and
tie points are used as seed points. The threshold of the normalised correlation
coefficient is set to a relatively low value (usually 0Æ7).
The approximate matching results are then refined by MPGC. This procedure is
performed on the original images. In order to solve the problems caused by
differences in image scales and rotations, the initial values of the reshaping
parameters in MPGC matching can be predetermined by using the collinearity
conditions for the four corner points of the image patches. The corresponding rays of
these four corners in the template image should intersect in object space and their
object coordinates can be determined (Fig. 13, right). Through the collinearity
equations the corresponding image coordinates in the search image can be
determined. The initial values of the four reshaping parameters are determined from
these four points and their correspondences.
Image Measurement with VirtuoZo Digital Photogrammetric System
A commercial photogrammetric package was also tested to recover the 3D model
of the Buddha statue. The matching method used by VirtuoZo is a global image
matching technique based on a relaxation algorithm (VirtuoZo NT, 1999). It uses both
grid point matching and feature point matching. The important aspect of this matching
algorithm is its smoothness constraint satisfaction procedure. With the smoothness
constraints, poor texture areas can be bridged, assuming that the model surface varies
smoothly over the image area. Through the VirtuoZo pre-processing module the user
can manually or semi-automatically measure some features like ridges, edges and
regions in difficult or hidden areas. These features are used as breaklines and planar
surfaces can be interpolated, for example, between two edges. In VirtuoZo, the feature-
point-based matching method is first used to compute a relative orientation between the
images of a stereopair. Then the measured features are used to weight the smoothness
constraints while the approximations found are used in the following global matching
method (Zhang et al., 1992).
Results from the Internet Images
Photo-triangulation
For every image found on the Internet, the pixel size and a focal length are
assumed, as well as the principal point, fixed in the centre of the images. With this last
assumption, the size of each image is considered to represent the original dimensions
of the photo, whereas in fact it could be just a part of an originally larger image. The
assumed pixel sizes are between 0Æ03 and 0Æ05 mm.
As no other information is available, an interactive determination of the camera
positions was first performed, using a graphical user interface program, also varying
the value of the focal length and using some control points measured on Professor
Kostka’s contour plot. Then these approximations were refined with a single photo
spatial resection solution.
The final orientation parameters were then recovered with a bundle adjustment.
The image correspondences of the tie points were obtained semi-automatically with
adaptive least squares matching (Gru¨n, 1985). The final average standard deviations of
Gru
¨net al. Photogrammetric reconstruction of the Great Buddha of Bamiyan, Afghanistan
188 2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd.
the object point coordinates located on the Buddha itself and on its immediate vicinity
are r
x,y
¼0Æ13 m and r
z
¼0Æ30 m. The recovered camera poses and the tie and control
points used are shown in Fig. 14, left.
Image Measurement and Surface Reconstruction with the Matching Algorithm
For the surface reconstruction, due to the scale and rotation differences among
the images, the modified version of the developed MPGC matching algorithm was
applied. A point cloud of approximately 6000 points was obtained. Some gaps are
present in the results (Fig. 14, centre) because of surface changes due to the
different times of image acquisition and to the low texture in some areas. For the
conversion of the point cloud to a triangular surface mesh, a 2Æ5D Delaunay
triangulation is applied, which is clearly sub-optimal. The textured 3D model is
shown in Fig. 14, right.
Results from the Tourist Images
Photo-triangulation
The pixel size was provided by Harald Baumgartner who scanned the original
images (35 mm film diapositive slides). He also provided three different values of focal
length. Therefore, a raw orientation could be performed using some control points
measured on Professor Kostka’s contour plot and a spatial resection algorithm.
The final orientation parameters were recovered with a bundle adjustment
(Fig. 15, left). The final average standard deviations of the object point coordinates
located on the Buddha itself and on its immediate vicinity were r
x,y
¼0Æ12 m and
r
z
¼0Æ17 m.
Image Measurement and Surface Reconstruction with the Matching Algorithm
The modified matching procedure on the tourist images resulted in 5585 points.
Some blunders were deleted by manual editing and the final point cloud and the related
textured model are shown in Fig. 15, centre and right. The relatively low image
resolution of the tourist data-set results in a coarse but quite complete 3D model.
Results from the Metric Images
Photo-triangulation
In all TAF images the principal point (PP) is defined as the intersection of the
straight line joining the two fiducial marks on the upper and lower side of the image
and the horizontal line passing through the horizontal index defined on the right side of
the image (Fig. 16, centre). The focal length of the camera (160Æ29 mm) is given in
Kostka (1974), where the acquisition procedure is also described (Fig. 16, right). The
images were acquired in the normal case, with a double baseline and at a distance of
about 130 to 150 m from the statue (Fig. 16, right).
Using this information and some control points measured on the contour plot, the
first approximations of the exterior orientation were achieved.
The Photogrammetric Record
2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd. 189
With the following bundle adjustment all the parameters of the three images were
recovered (Fig. 17); the final standard deviations of the object point coordinates
located on the Buddha itself and in its immediate vicinity were r
x,y
¼0Æ07 m and r
z
¼
0Æ14 m. These high values can be explained by the unfavourable control point
distribution covering just a small part of the images (see Fig. 14, bigger dots).
Image Measurement and Surface Reconstruction with the Matching Algorithm
The MPGC matching algorithm resulted in fairly reliable and precise results.
From the three metric images, 49 333 points (without the surrounding rocks) and
73 640 points (with part of the surrounding rocks) were obtained. The point cloud data
is shown in Fig. 18 (centre), as well as a textured 3D model (right). Although an
automatic blunder and occlusion detection process was used, some blunders are
present in the 3D point cloud and they were removed with manual editing. As shown
Fig. 16. The TAF camera (left) and the interior orientation of the images (centre). The acquisition
geometry of the three metric images as presented in Kostka (1974).
Fig. 15. A view of the recovered camera poses of the tourist images (left). Point cloud (centre) and
textured 3D model, obtained with ETH Zu¨ rich’s automated matching program.
Gru
¨net al. Photogrammetric reconstruction of the Great Buddha of Bamiyan, Afghanistan
190 2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd.
in Fig. 18, there are some gaps in the point cloud, mainly due to the shading effects
caused by the variation of the illumination conditions during the image acquisition
(typical areas are marked in Fig. 18).
As shown in Fig. 19, the dress of the Buddha was rich in folds, which were
between 5 and 15 cm in width. Many folds could not be exactly reconstructed with the
automatic program. Fig. 19 also shows some of the problems that the automated
matcher had to deal with: high noise values in the shadow areas, large differences
between illuminated and shadow areas, and artefacts (straight lines) from the
photographic development process. Other reasons for failure are: the image patches of
least squares matching are assumed to correspond to planar object surface patches but,
along the folds, this assumption is no longer valid and the small features are smoothed
Fig. 17. The positions of the three metric images recovered after the bundle adjustment (left) and the
position of the control points on the statue (right).
Fig. 18. Left: Two original metric images of the Great Buddha (part). Centre: 3D point cloud generated
by automatic matching on the metric images. Right: Textured 3D model.
The Photogrammetric Record
2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd. 191
out (Fig. 20). Secondly, taking a very small patch size could theoretically avoid or
reduce the smoothing effect, but may result in problems with the determination of the
reshaping parameters. Fig. 20 compares the cases of small and large patch sizes over
the folds of the dress. While a large patch tends to smooth out the local 3D
information, a small patch may not include sufficient signal content for the matching.
In future work, edge-based matching will also be explored to reconstruct those very
small features of the Buddha statue.
In view of these matching problems precise manual measurements were used to
reconstruct the exact shape of the dress and other very small features of the statue in
order to provide complete data for a future physical reconstruction.
Image Measurement and Surface Reconstruction with a Commercial Package
It was decided to test also the VirtuoZo digital photogrammetric system for
comparison. For the matching procedure, see the details already given on the VirtuoZo
system.
In the application, the metric images of Fig. 10(b) and (c) from Professor Kostka
were used to reconstruct the 3D model. A regular image grid with 9 pixels spacing was
matched using a patch size of 9 ·9 pixels and four pyramid levels. As a result, a very
dense point cloud of approximately 178 000 points was generated (Fig. 21). The statue
as well as the rock around it is well reconstructed, but due to the smoothness constraint
Fig. 19. The folds on the dress that the automated algorithms cannot completely reconstruct (left).
Corresponding left and right image patches, showing large differences in image content (right).
Fig. 20. Patch definition in least squares matching. p
s
,p
l
represent the image patches (small, large). r
s
,
r
l
represent the object planar patches (small, large).
Gru
¨net al. Photogrammetric reconstruction of the Great Buddha of Bamiyan, Afghanistan
192 2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd.
and grid-point-based matching the small folds on the body of the Buddha were filtered
or skipped and they are not visible. A 2Æ5D Delaunay triangulation was applied for the
surface modelling. Without losing its topology, the 3D surface model of the Buddha
was expanded to a plane by transforming the Cartesian coordinate system to a
cylindrical coordinate frame. In the defined qhf cylinder frame, frefers to the vertical
cylinder axis crossing the model centre and parallel to the original Yaxis of the
Cartesian object coordinate system; qis the Euclidean distance from the surface point
to the zaxis and his the angle around the zaxis. The 2Æ5D triangulation was done in
the hf plane and the final shaded model of the triangulated mesh is shown in Fig. 21,
right. The shaded model looks ‘‘bumpy’’, mainly due to small measurement errors and
inconsistencies in surface modelling.
The central image of the metric data-set was mapped onto the 3D geometric
surface to achieve a photorealistic virtual model (Fig. 22). The lower parts of the legs
are not completely modelled because in the stereomodel used the legs were not visible.
Fig. 21. The Great Buddha generated with VirtuoZo matching on the metric images (left: point cloud,
right: shaded model).
Fig. 22. Visualisation of 3D model of the Great Buddha in textured mode.
The Photogrammetric Record
2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd. 193
Image Measurement and Surface Reconstruction with Manual Measurements
None of the automated procedures presented could recover the small details of the
dress, therefore only precise manual measurements could reconstruct the exact shape
and curvature of the folds. For this reason, the metric images were imported into the
VirtuoZo stereo digitise module (VirtuoZo NT, 1999) and manual stereoscopic
measurements were performed. Three stereomodels were set up and points were
measured along horizontal profiles at 20 cm intervals, while the folds and the main
edges were measured as breaklines.
In the point visualisation of Fig. 23 it is already possible to distinguish the shapes
of the folds on the dress. This point cloud is not dense enough (except in the area of the
folds) to generate a complete triangulation mesh with commercial reverse engineering
software. Therefore, the generation of the surface is performed with the 2Æ5D Delaunay
method, by dividing the measured clouds into separate parts. A mesh is created for
each single point cloud and then all the surfaces are merged together with Geomagic
Studio software (Geomagic, 2004). The folds of the dress are now well reconstructed
and modelled, as shown in Fig. 24.
Fig. 23. The point cloud of the manual measurement. The main edges and the structures of the folds,
measured as breaklines, are now clearly visible.
Fig. 24. Visualisation in wireframe mode of the reconstructed 3D structures on the central part of the
dress of the Buddha (left). Comparison with the original structures (right).
Gru
¨net al. Photogrammetric reconstruction of the Great Buddha of Bamiyan, Afghanistan
194 2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd.
With Geomagic Studio some editing operations on the meshes are also performed:
(a) hole filling: polygon gaps are filled by constructing triangular structures,
respecting the surrounding area;
(b) noise reduction: spikes are removed with smoothing functions;
(c) edge correction: faces are split (divided into two parts), moved to another
location or contracted;
Fig. 25. Shaded model of the Buddha, reconstructed with manual measurements (left) and the related
textured 3D model (right).
Fig. 26. Anaglyph images of the reconstructed 3D model.
The Photogrammetric Record
2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd. 195
(d) polygon reduction: in some areas, the number of triangles is reduced, pre-
serving the shape of the object.
The final 3D model, displayed in Fig. 25, shows also the reconstructed folds of
the dress. Compared to Fig. 21 (right), this represents a much better result.
For photorealistic visualisation, the central image of the metric data-set is mapped
onto the model, as shown in Fig. 25 (right).
3D Model Visualisation
Many different tools are available to display 3D models, shareware or commercial
software, with or without real-time performance, interactive or not.
The generated model can be visualised with software developed at ETH Zu¨rich,
called Disp3D (Buehrer et al., 2001; Gru¨n et al., 2001). It allows the visualisation of a
3D model as point cloud, in shaded or textured mode, as well as with interactive
navigation.
One of the few portable formats to interactively display a 3D model like the
reconstructed Buddha statue is VRML. With free packages like Cosmo Player or
Vrweb it is possible to display the model and navigate through it or to fly along
predefined paths.
Computer animation software (for example, Maya) is generally used to create
high quality animations of 3D models. Some examples are presented by ETH Zu¨rich at
http://www.photogrammetry.ethz.ch/research/bamiyan/buddha/animations.html. Maya
renders the model offline, using anti-aliasing functions and producing portable videos
like MPEG or AVI.
Finally, a further way of displaying a static view of 3D models is based on
traditional anaglyph images (Fig. 26) in which a stereoscopic view is generated using
the complementarity of colours in the RGB channels. The depth information within
each model can then be viewed by means of the coloured filter spectacles which may
be assumed still to be present in the toolkits of most readers of The Photogrammetric
Record.
The Physical Reconstruction of the Great Buddha at 1:200 Scale
The 3D computer model reconstructed with the manual procedure was used for a
physical reconstruction of the Great Buddha (Fig. 27). At the Institute of Machine
Tools and Production, ETH Zu¨ rich, R. Zanini and J. Wirth have recreated a 1:200
model statue of the Great Buddha (approximately 25 cm high).
For this purpose, the point cloud of the photogrammetric reconstruction was
imported to a digitally programmed machine tool (Starrag NF100) without any further
processing (Wirth, 2002). The machine works on polyurethane boxes and follows
milling paths calculated directly from the point cloud. The physical model is created in
three steps: (1) a roughing path, (2) a pre-smoothing path and (3) the final smoothing
path.
The time needed for preparing the production data was about 2 h while the milling
of the statue was done in about 8 h.
Gru
¨net al. Photogrammetric reconstruction of the Great Buddha of Bamiyan, Afghanistan
196 2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd.
Conclusions and Future Work
It has been shown that even fairly complex structures can be reconstructed
photogrammetrically from Internet images, without any prior knowledge about the
geometry of those images. This 3D modelling can be done even automatically with an
advanced matching strategy. However, in such reconstructed 3D models essential small
features, like the folds of the dress and some important edges, are missed. For the
generation of a complete and detailed model, manual photogrammetric measurements
are indispensable. Of course they can be performed in stereomodels composed of
digital images.
Beyond the issue of measurement, a major problem still exists with the surface
modelling. ETH Zu¨rich’s own software and commercial modelling software were not
satisfactory and finally more time was spent on modelling than on measurement. In
order to have a smooth production process the measurement procedure should be
adapted to the capabilities of the 3D modeller. This was not the case in the project and
this necessarily tighter connection was learned the hard way.
The final 3D model of the Great Buddha can serve as the basis for the physical
reconstruction. The most recent developments call for the placement of a 5Æ3 m high
model in the National Museum of Kabul, Afghanistan.
The generation of this computer model concludes the first part of the work on the
Bamiyan project. During the last week of August 2003, 4 days were spent in a field
campaign on site. Photos were taken with various cameras (Rollei 6006 metric, Sony
Cybershot still video) of the complete rock face (more than 1 km in length, up to 100 m
high) and of the back-walls of the Buddha caves (to record the situation after the
demolition).
An extended 3D model of the site will be produced using SPOT 5 images,
including an area of about 11 ·18 km
2
of DTM with Ikonos texture, the rock face at
higher resolution and the Buddhas themselves at very high resolution. This requires
also the photogrammetric reconstruction of the Small Buddha, which is currently
underway. Furthermore, it is planned to reinsert the destroyed frescos into the complex
model, using old imagery as well.
Various visualisations and animations are planned, some of them contributing to a
90 min film about ‘‘The Great Buddhas’’, which is planned to be shown in cinemas and
on TV in 2005.
Acknowledgements
The authors would like to thank all the people who took part in this project and
in particular: Professor Kostka, Technical University of Graz, for the three metric
images; Vexcel Inc., for scanning the metric images; B. Weber, Founder of the
New7Wonders Society; P. Bucherer, Director of the Afghanistan Museum in
Bubendorf, Switzerland; Yuguang Li for the manual measurements on the metric
images; Robert Zanini, Joachim Wirth and the Institute of Machine Tools Production,
ETH Zurich, for the physical reconstruction of the statue at scale 1:200; Tom Bilson,
Courtauld Institute of Art, London, for some Internet images of the Bamiyan statues;
Harald Baumgartner for the ‘‘tourist’’ images; all the websites where images of the
statues were found.
The Photogrammetric Record
2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd. 197
references
AIIS, 2004. The American Institute of Indian Studies’ photo archive. http://dsal.uchicago.edu/images/aiis/
bamiyan/ [Accessed: 7th May 2004].
Baltsavias,E. P., 1991. Multiphoto Geometrically Constrained Matching. Dissertation No. 9561, ETH
Zu¨ rich. 221 pages.
Buehrer,T.,Zhang,L.,Gru
¨n,A.,Fraser,C. and Ruther,H., 2001. Photogrammetric recon-
struction and 3D visualisation of Bet Giorgis, a rock-hewn church in Ethiopia. Proceedings of the
18th CIPA Symposium, ‘‘Surveying and Documentation of Historic Sites’, Potsdam, Germany. 754
pages: 338–344.
Byron,R., 1981 [1937]. The Road to Oxiana. Picador, London. 286 pages.
Finsterwalder,S., 1896. Zur photogrammetrischen Praxis. Zeitschrift fu¨ r Vermessungswesen, 225–240.
Finsterwalder,S. and Hofmann,W., 1968. Photogrammetrie. de Gruyter Lehrbuch, Berlin. 455
pages: 119–120.
Geomagic, 2004. http://www.geomagic.com/products/studio [Accessed: 7th May 2004].
Gru
¨n,A., 1985. Adaptive least squares correlation: a powerful image matching technique. South African
Journal of Photogrammetry, Remote Sensing and Cartography, 14(3): 175–187.
Gru
¨n,A. and Baltsavias,E., 1988. Geometrically constrained multiphoto matching. Photogrammetric
Engineering and Remote Sensing, 54(5): 633–641.
Gru
¨n,A.,Zhang,L. and Visnovcova,J., 2001. Automatic reconstruction and visualization of a
complex Buddha tower of Bayon, Angkor, Cambodia. Proceedings of 21st Wissenschaftlich Tech-
nische Jahrestagung der Deutschen Gesellschaft fu¨ r Photogrammetrie und Fernerkundung (DGPF),
Konstanz, Germany, 4th to 7th September. 588 pages: 289–301.
Gru
¨n,A.,Remondino,F. and Zhang,L., 2003. Image-based Automatic Reconstruction of the Great
Buddha of Bamiyan, Afghanistan. Computer Vision and Pattern Recognition (CVPR) Workshop on
‘‘Application of Computer Vision in Archaeology’’, Madison, USA (on CD-ROM).
Hackin,J., 1928. The colossal Buddhas at Bamiyan. Their influence on Buddhist sculpture. Eastern Art,
1(2): 109–116.
ICOMOS, 2001. Press release. http://www.afghan-network.net/News/Archives/2001/Press-icomos.html
[Accessed: 6th May 2004].
IRAO, 2004. Istituto di Ricerca Arte Orientale. http://www.arteorientale.com [Accessed: March 2004].
Knobloch,E., 2002. The Archaeology and Architectureof Afghanistan. Tempus, Stroud. 192 pages: 88–95.
Kostka,R., 1974. Die stereophotogrammetrische Aufnahme des Grossen Buddha in Bamiyan.
Afghanistan Journal, 3(1): 65–74.
Muse
´e National des Arts Asiatiques-Guimet, 2004. http://www.museeguimet.fr [Accessed: March
2004].
Fig. 27. The milling machine used for the physical reconstruction of the Bamiyan Buddha (left) and an
image of the 1:200 model (right).
Gru
¨net al. Photogrammetric reconstruction of the Great Buddha of Bamiyan, Afghanistan
198 2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd.
UNESCO, 2004. http://www.unesco.org [Accessed: 6th May 2004].
VirtuoZo NT, 1999. Version 3.1 Manual, Supresoft Inc. 239 pages.
Wirth,J., 2002. Rapid Modeling. Gegensta¨ ndliches CAD fu¨ r die ‘‘begreifbare’’ Produktgestaltung.
Hanser, Munich. 220 pages.
Zhang,Z.,Zhang,J.,Wu,X. and Zhang,H., 1992. Global image matching with relaxation method.
Proceedings of the International Colloquium on Photogrammetry, Remote Sensing and Geographic
Information Systems, Wuhan, China, 11th to 14th May: 175–188.
Re´sume´
Il y a presque 1700 ans, dans la valle´ e de Bamiyan, en Afghanistan, deux
grandes statues du Bouddha ont e´te´ sculpte´ es dans la roche se´dimentaire de la
re´gion. Elles e´taient hautes de 53 et 38 me`tres respectivement; la statue la plus
haute e´tait conside´re´e comme la plus grande repre´ sentation d’un Bouddha
dans le monde entier. En mars 2001 la milice des Talibans a de´moli ces statues
colossales. Apre`s leur destruction, notre groupe a exe´cute´ une reconstruction a
l’ordinateur du grand Bouddha, qui peut servir de base pour sa reconstruction
physique. Dans cet article nous de´crivons les re´sultats obtenus avec la
reconstruction en trois dimensions (3D) de la statue, sous forme d’images; la
reconstruction a e´te´ exe´ cute´e avec trois diffe´rents groupes d’images et en
employant diffe´rents algorithmes et techniques photogramme´triques.
Zusammenfassung
Im Tal von Bamiyan, Afghanistan, wurden vor ca. 1700 Jahren zwei
große Buddhastatuen aus einer la¨ ngeren Felswand herausgearbeitet,
zusammen mit Hunderten von Ho¨hlen und Grotten, welche von buddhis-
tischen Mo¨nchen genutzt wurden. Mit 53 bzw. 38 Metern Ho¨he za¨ hlten diese
Statuen zu den gro¨ ßten stehenden Buddhastatuen der Welt. Im Ma¨rz 2001
zersto¨rten Talibanmilizen die kolossalen Figuren. Nach der Zersto¨ rung fu¨ hrte
unsere Gruppe die Computerrekonstruktion des Großen Buddha auf der
Basis von fru¨heren Bildern durch. Diese 3D Rekonstruktion kann als
Grundlage fu¨r eine spa¨tere physische Rekonstruktion benutzt werden.
In diesem Beitrag berichten wir u¨ ber die Resultate der 3D Rekonstruk-
tion, welche mit drei unterschiedlichen Bilddatensa¨ tzen (Amateuraufnahmen
und Messbilder) und mit verschiedenen photogrammetrischen Techniken und
Algorithmen (manuell und automatisch) durchgefu¨ hrt wurde.
Resumen
Hace aproximadamente unos 1700 an˜ os se excavaron en el valle de
Bamiyan, Afganistan, dos grandes esculturas de Buda de pie en la roca
sedimentaria propia de la regio´ n. Tenı´an 53 y 38 metros de alto y la mayor
era la representacio´n ma´s alta del mundo de una figura de Buda de pie. En
marzo de 2001 el gobierno taliba´n hizo derribar las colosales esculturas.
Tras la destruccio´ n nuestro equipo hizo una reconstruccio´ n digital del Gran
Buda que puede servir de base para la reconstruccio´n fı´sica. Este artı´culo
describe los resultados de la reconstruccio´ n en 3D de la escultura a partir
de tres conjuntos distintos de ima´genes, utilizando diferentes te´cnicas
fotograme´tricas y algoritmos.
The Photogrammetric Record
2004 The Remote Sensing and Photogrammetry Society and Blackwell Publishing Ltd. 199
... Imagery can be obtained from various cameras ranging from large format metric cameras, where high accuracy and archival value are essential, to smaller format 35mm cameras which are helpful where access is restricted. Compared to standard aerial (orbital) photogrammetry, close-range photogrammetry collects images taken at a closer range [14,15]. One major drawback of photographic reconstruction is the need for post-processing. ...
Article
The recent emergence of technology as digital tools has provided architects with possibilities for 3D modelling and simulation of built environments. Currently, the "digital twin" and Building Information Management (BIM) techniques enable analytical tools like connection analysis and energy performance modelling in tandem with the ideation of realistic 3D illustrations. These methods are now seen emerging in managing and documenting heritage structures. A Heritage Building Information Management (HBIM) platform is a multi-disciplinary process and a valuable tool to automate, manage, and document heritage structures. However, making the final product in an HBIM platform involves several laborious steps of processing, segmenting, and integrating raw data from different sources, such as laser-scanned point data and image data. Hence, an expedited, streamlined platform is necessary to facilitate conservation architects and other agencies involved in documenting, maintaining and interpreting heritage buildings and structures. Newer technologies like Artificial Intelligence can enhance previously unavailable opportunities to automate specific tasks and uncover paradigms through predictable simulations. For this, the researchers reviewed the literature on two fronts: i)Investigating process flow and identifying the issues and challenges in gathering digital data of the existing building effectively. ii)Reviewing the literature to understand the models and tech solutions that revolutionized building heritage building information models for practical use. This study unravels workflow patterns and challenges based on these two sections of this literature review. Further, this research explores and analyses the trends and techniques employed in a typical heritage Building Information Management (HBIM) platform. The focus is on AI's capabilities as a digital tool in different stages of documentation and interpretation, performance measurement and simulation of heritage buildings, such as energy performance modeling, 4D (3D+Time) for projects planning, cost solutions, and simulating the historical significance to propagate the importance of HBIM. This study referred to the progress in the scientific literature on HBIM after the initial studies of C. Dore, and M. Murphy (2009).
... Three-dimensional (3D) models created from archived photographs can aid in remote analysis and visualization, reducing the need for invasive physical interventions. Additionally, in cases of asset destruction, a digital copy must be created, as seen in the case of the Buddhas of Bamiyan [4]. In fact, three-dimensional imaging technology has become a significant tool in the preservation and restoration of cultural heritage assets. ...
Article
Full-text available
The goal of this research is to make photogrammetric surveys of the walls of Cortona from 2012 accessible using new methodologies for recovering photographic material. This will allow a team of archaeologists to carry out a virtual reconnaissance of the surveyed stretch of wall as well as provide the basis for future investigations into any potential changes that may have occurred in the wall since 2012. Photogrammetry is a widely used technique in archaeology that can help researchers accurately measure, reconstruct, and analyze different architectural components of the wall. By using state-of-the-art photogrammetric techniques, including advanced computer vision algorithms, our team aims to produce high-quality 3D models and accurate measurements of different parts of the wall. The results of this research project will enable archaeologists to gain a more comprehensive understanding of the layout of the fortifications and the role of the Cortonese walls in the historical context of the area. Additionally, the research project will provide a detailed documentation of the wall that will be useful for both archaeological researchers and cultural heritage organizations. Finally, the research project will also provide the basis for future investigations into potential changes that may have occurred in the wall since 2012, which will be important for monitoring conservation and restoration efforts and providing an up-to-date record of the wall’s state of preservation.
Chapter
Full-text available
As this chapter shows, digital 3D reconstructions of historic architecture serve many purposes in research and related areas. This comprises answering research questions by creating a 3D model, preserving cultural heritage, communicating knowledge in education, and providing a structure for knowledge organization. The process of creating a 3D reconstruction is often challenging, for example, because of lacking or ambiguous sources. In order to create a 3D reconstruction based on scientific values, guidelines, and standards are needed.
Conference Paper
Full-text available
Image matching is the core of many computer vision applications for cultural heritage. The standard image matching pipeline detects keypoints at the beginning and freezes them until bundle adjustment, by which keypoints are allowed to move in order to improve the overall scene estimation. Recent deep image matching approaches do not follow this scheme, historically imposed by computational limits, and progressively refine the localization of the matches in a coarse-to-fine manner. This paper investigates the use of traditional computer vision approaches based on template matching to update the keypoint position throughout the whole matching pipeline. In order to improve the accuracy of the template matching, the usage of the coarse-to-fine refinement is explored and a novel normalization strategy for the local keypoint patches is designed. Specifically, the proposed patch normalization assumes a local piece-wise planar approximation of the scene and warps the corresponding patches according to a “middle homography”, so that, after normalization, patch distortion is roughly equally distributed within the two original patches. The experimental comparison of the considered approaches, mainly focused on cultural heritage scenes but straightforwardly generalizable to other common scenarios, shows the strengths and limitations of each evaluated method. This analysis indicates promising and interesting results for the investigated approaches, which can effectively be deployed to design better image matching solutions.
Article
Full-text available
The inventory and categorization of an extensive coastal boulder assemblage originating from storm wave transport on the coastline of Fenoliga Island (Northern Adriatic Sea, southern Istria, Croatia) are presented and discussed herein. The study adopted the use of a commercial Uncrewed Aerial Vehicle (UAV) and Structure from Motion-MultiView Stereo (SfM-MVS) photogrammetry for the construction of a 3D model of the island. A Digital Elevation Model (DEM) and an orthomosaic were produced and employed for the mapping of the boulder assemblage in a GIS. In total, 592 boulders were identified and mapped. Using SfM-MVS-derived products allowed for the identified boulders to be categorized based on size classification. Amassed data relating to the boulder characteristics was inserted and stored in a GIS, including the results of a comparative assessment with historical Google Earth imagery which enabled the ‘quantification of boulder transport over a 9-year timeframe’. Field evidence indicates that boulders were created in-situ via the quarrying of bedrock strata by breaking waves causing increased water pressure within preexisting surfaces of weakness such as bedding planes and sub-vertical fractures. Once detached, the boulders were transported and deposited during storm wave events. Repeated storm events can further displace previously detached clasts.
Chapter
Nowadays, billions of photographs are available on the internet, under an open-access/open-data status, thus forming the largest and most diverse collection of optical data ever assembled. These images can be easily recovered, and digitally processed for the photogrammetric restoration of monuments, partly or completely destroyed, thus rendering clear the effective contribution of crowdsourcing under the perspective of restoring our cultural past.This is the case of the lost landmark of Chairedin Pasha Sarai in Ioannina city, Greece: the extensive scientific research on the building complex’s location, form and dimensions aiming to the respective 3D digital model design and development, has been further creatively supported and supplemented by downloading from the internet and making use of several photographs depicting different aspects of the building, at various chronological periods, before its final complete destruction, thus allowing the visual 3D digital restoration of the landmark, as well as its interaction with the surrounding urban web and its successive alterations.In this context, the crowdsourcing dimension might be considered as a basic element, a catalyst, for the restoration of cultural heritage, while the active participation of society in matters of culture, as is the contribution in monuments’ documentation procedure, awakens the collective consciousness in the direction of the protection, preservation and highlighting of cultural heritage.KeywordsCrowdsourcingCultural HeritageDocumentationDigitizationGISPhotogrammetry3D ModellingOpen SourcesIoannina
Chapter
In 2015, we witnessed the horrific destruction of the Mosul Museum and other sites by the so-called Islamic State. Rekrei is a crowd-sourcing platform that was initiated in response to this destruction in hopes of being able to preserve the memory of lost heritage. The platform provides a venue for anyone to contribute photos, help organize collections, or even download and process those photos photogrammetrically. Within the first months of launching the project, the volunteer community produced some impressive results, and the project even partnered with the Economist to produce a virtual museum experience retelling the story of the Mosul Museum, the destruction, and the hope for a future. Fast forward a few years to 2019 when Rekrei was invited to participate in the first public exhibition in the museum (and the city) after the occupation by Daesh. This paper looks at the role of Rekrei in preserving the memory of lost heritage from the Mosul Cultural Museum, and finally bringing that heritage back home to Mosul and the role that heritage plays in helping a community to recover in a post-conflict situation.KeywordsCultural heritageMosul cultural museumRekrei
Article
Full-text available
Areas suffering territorial imbalances are at risk, as climate change and inherent social-spatial vulnerabilities bring uncertainty over their capacity to achieve sustainable development. To enable policy-makers' vision for them, the paper proposes adopting digitally-enhanced disaster risk reduction practices which promote civic engagement and evidence-based decision-making. The study introduces the concept of Territorial Digital Twins (TDTs) demonstrating the potential benefits of networking distributed information resources and enabling their integration in two paradigmatic Italian inner mountain areas, which are linked to different priorities of the Sendai Framework for Disaster Risk Reduction 2015–2030 (SFDRR). This contribution argues the importance of complementing the technical approach with the social perspective in the development of TDTs oriented at increasing the community resilience of territorially imbalanced areas. It then illustrates the usefulness of exploiting TDTs (from photogrammetry, GIS mapping, Space Syntax analysis) to overcome the barriers to the SFDRR, and enable its effective implementation in localities.
Article
Full-text available
This paper investigates the potential of dense multi-image 3D photogrammetric reconstruction of destroyed cultural heritage monuments by employing public domain imagery for heritage site visitors. This work focuses on the digital reconstruction of the Temple of Bel, one of the heritage monuments in Palmyra, Syria, which was demolished in the summer of 2015 due to armed conflict. This temple is believed to be one of the most significant religious structures of the first century AD in the Middle East and North Africa (MENA) region with its unique design and condition before destruction actions. The process is carried out using solely one source of images; the freely available visitors’ images collected from the social media platforms and web search engines. This paper presents a digital 3D reconstruction workflow for the collected images using an advanced photogrammetry pipeline and dense image matching software. The virtually reconstructed outputs will be managed and implemented efficiently in Unity3D to create an entire 3D virtual interactive environment for the deconstructed temple to be visualised and experienced using the new Oculus Quest VR headset. The virtual Palmyra’s visitor will be offered an enhanced walk-through off-site interactive, immersive experience compared to the real-world one, which is non-existing and unobtainable at the site in the current time.
Article
Full-text available
ABSTRACT The Adaptive Least Squares Correlation is a very potent and flexible technique for all kinds of data matching problems. Here its application to image matching is outlined. It allows for simultaneous radiometric corrections and local geometrical image shaping, whereby the system parameters are automatically assessed, corrected, and thus optimized during the least squares iterations. The various tools of least squares estimation can be favourably utilized for the assessment of the correlation quality. Furthermore, the system allows for stabilization and improvement of the correlation procedure through the simultaneous consideration of geometrical constraints, e.g. the collinearity condition. Some exciting new perspectives are emphasized, as for example multiphoto correlation, multitemporal and multisensor correlation, multipoint correlation, and simultaneous correlation/triangulation.
Article
Full-text available
The Adaptive Least-Squares Correlation, combining gray level matching with geometrical constraints, is applied for X, Y, Z object coordinate determination. The constraints used are the collinearity conditions. A new aspect is the simultaneous use of multiple (more than two) scenes. This paper outlines the mathematical model and highlights some essential features of the algorithm with practical data. Tests using CCD camera data in a close-range environment were performed on the aspects of pull-in range, occlusions, and reliability (multiple solutions, mismatch). In all cases remarkable advantages result from the use of geometrical constraints (conditional one-dimensional search) and multiple scenes. Depth errors of 5% average depth (d//o) (6 pixels pull-in range) and 10% d//o (12 pixels pull-in range) were examined, with 100 percent and 70 percent success rate, respectively.
Article
Full-text available
The Hindu and Buddhist monuments of Angkor in Cambodia are the legacy of highly developed Khmer empires. Well-known for their structural and surface complexity they constitute a great challenge to any attempt towards precise and detailed 3D measurements and modeling. This paper reports about a pilot project using modern techniques of analytical and digital photogrammetry to derive a photorealistic 3D model of one of the very complex towers of the famous Bayon temple of the ancient city of Angkor Thom. This high quality model will then be subject to visualization and animation. On occasion of a balloon photogrammetry mission over the Bayon temple of Angkor, the first author took a number of tourist-type terrestrial images with a Minolta Dynax 500si analogue SRL camera of one of the many Buddha-faced towers of Bayon. We aim at deriving automatically, after scanning of the images, a texture mapped 3D model of the very complex object. In a first step we have generated already such a model with a mixture of manual (phototriangulation) and automated procedures (image matching for surface reconstruction, editing for blunder removal, texture mapping, visualization and animation). This result has been presented in Visnovcova et al., 2001 and includes already
Conference Paper
Full-text available
In the great valley of Bamiyan, Afghanistan, two big standing Buddha statues were carved out of the sedimentary rock of the region around the second to fourth centuries AD. The larger statue was 53 meters high while the smaller Buddha measured 35 m. The two colossal statues were demolished on March 2001 by the Taleban, using mortars, dynamite and rockets. After the destruction, a consortium was founded to rebuild the Great Buddha at original shape, size and place. Our group performed the required computer reconstruction, which can serve as the basis for the physical reconstruction. The work has been done using three different types of imagery in parallel and in this paper we present our results of the 3D computer reconstruction of the statue using automated approaches of image analysis.
Article
3., völlig neu bearb. Aufl
Press release. http://www.afghan-network
ICOMOS, 2001. Press release. http://www.afghan-network.net/News/Archives/2001/Press-icomos.html [Accessed: 6th May 2004].
The colossal Buddhas at Bamiyan
  • J Hackin
Hackin, J., 1928. The colossal Buddhas at Bamiyan. Their influence on Buddhist sculpture. Eastern Art, 1(2): 109-116.