DISTANCE-TRAINING FOR IMAGE-BASED 3D MODELLING OF ARCHEOLOGICAL SITES IN REMOTE REGIONS

Abstract

The impressive success of Structure-from-Motion Photogrammetry (SfM) has spread out the application of image-based 3D reconstruction to a larger community. In the field of Archeological Heritage documentation, this has opened the possibility of training local people to accomplish photogrammetric data acquisition in those remote regions where the organization of 3D surveying missions from outside may be difficult, costly or even impossible. On one side, SfM along with low-cost cameras makes this solution viable. On the other, the achievement of high-quality photogrammetric outputs requires a correct image acquisition stage, being this the only stage that necessarily has to be accomplished locally. This paper starts from the analysis of the well-know “3×3 Rules” proposed in 1994 when photogrammetry with amateur camera was the state-of-the art approach and revises those guidelines to adapt to SfM. Three aspects of data acquisition are considered: geometry (control information, photogrammetric network), imaging (camera/lens selection and setup, illumination), and organization. These guidelines are compared to a real case study focused on Ziggurat Chogha Zanbil (Iran), where four blocks from ground stations and drone were collected with the purpose of 3D modelling.

Full text

DISTANCE-TRAINING FOR IMAGE-BASED 3D MODELLING OF ARCHEOLOGICAL
SITES IN REMOTE REGIONS
V. Yordanov1,2, A. Mostafavi3, M. Scaioni4
1 Vasil Levski National Military University, Veliko Tarnovo, Bulgaria
2 Dept. of Civil and Environmental Engineering (DICA)
Politecnico di Milano, Piazza Leonardo da Vinci 32, Milan, Italy – email: vasil.yordanov@polimi.it
3 Node-Office Tehran, Iran – email: armin.mstfv@gmail.com
4 Dept. of Architecture, Built environment and Construction engineering (DABC)
Politecnico di Milano, Via Ponzio 31, Milan, Italy – email: marco.scaioni@polimi.it
KEY WORDS: Archeology, Photogrammetry, Remote Areas, Structure-from-Motion, Training
ABSTRACT:
The impressive success of Structure-from-Motion Photogrammetry (SfM) has spread out the application of image-based 3D
reconstruction to a larger community. In the field of Archeological Heritage documentation, this has opened the possibility of
training local people to accomplish photogrammetric data acquisition in those remote regions where the organization of 3D
surveying missions from outside may be difficult, costly or even impossible. On one side, SfM along with low-cost cameras makes
this solution viable. On the other, the achievement of high-quality photogrammetric outputs requires a correct image acquisition
stage, being this the only stage that necessarily has to be accomplished locally. This paper starts from the analysis of the well-know
“3x3 Rules” proposed in 1994 when photogrammetry with amateur camera was the state-of-the art approach and revises those
guidelines to adapt to SfM. Three aspects of data acquisition are considered: geometry (control information, photogrammetric
network), imaging (camera/lens selection and setup, illumination), and organization. These guidelines are compared to a real case
study focused on Ziggurat Chogha Zanbil (Iran), where four blocks from ground stations and drone were collected with the purpose
of 3D modelling.
1. INTRODUCTION
The vaste Archeological Heritage on Earth may take great
advantage from existing technology for 3D surveying and
modelling, which play a paramount role in digital archiving,
restoration, dissemination, and communication. In the past the
photogrammetric surveying relied on the use of specific
metric/semimetric cameras for data acquisition, and the use of
complex analytical or digital procedure for extracting 2D and
3D information (such as plans, prospects, cross-sections, and
orthophotos/rectified photos). On one side, the diffusion of
digital non-metric cameras (see Waldhäusl and Ogleby, 1994)
started to allow a dramatic cost reduction process in
photogrammetric data acquisition. Thanks to rigorous but
simple procedures, imaging sensors could be calibrated to
obtain accurate metric outputs (Luhmann et al., 2016). On the
other side, a progressive development of digital
photogrammetry has undergone along with the impressive
improvement of computing power of standard computers. But
the most relevant step forward in image-based 3D modelling
was the success of the so-called Structure-from-Motion
Photogrammetry (in the sequel simply SfM). In a recent
editorial, Granshaw (2018b) reviewed the origin of this
technique that found its roots in both Photogrammetry
(Luhmann et al., 2014) and Computer Vision (Hartley and
Zisserman, 2006) domains. Originally, the term Structure-from-
Motion only referred to the image orientation phase: at the very
beginning considering the geometric model (Ullman, 1979),
then including the automatic search for corresponding points
using image matching (Snavely et al., 2006). In the last ten
years, the popularity gained by this technique among non-
specialists, coupled with improvements in dense surface
matching (Gruen, 2012), led to extend the term to cover both
phases of image-based 3D reconstruction process.
By combining automatic image processing algorithms for image
registration, which are robust against the use of convergent
images and radiometric changes typical of close-range
photogrammetric blocks, bundle adjustment (including self-
calibration) and dense surface matching techniques, SfM
provides the users with an automatic pipeline to obtain efficient
3D reconstructions from images. This success is also motivated
by the diffusion of powerful, easy-to-use and low-cost (or open
source) software packages, which implement SfM in efficient
way. After a few years when terrestrial laser scanning (TLS)
sensors seemed to be the uncontrasted tools for 3D point cloud
acquisition, SfM has granted again the photogrammetry as one
of the leading techniques to this purpose. If compared to TLS,
SfM has also the advantage of a much lower cost, in particular
for purchasing the necessary hardware.
Today SfM is widely applied in many domains, including
Cultural Heritage (CH). Thanks to the economic sustainability
and the apparent simplicity in its usage, SfM may be also
operated by non-experts to survey archeological sites located in
remote areas where specific surveying campaigns cannot be
organized (see, e.g., Barazzetti et al., 2011). This includes the
case of countries affected by war events or characterized by
local unstable social/political condition, preventing experts to
come from outside the region to carry out 3D surveying of the
CH. The use of amateur digital cameras, the availability of
cheap small drones (see Granshaw, 2018a), and the chance of
using low-cost photogrammetric packages, can be all together
exploited by local people to do the 3D surveying operations.
On the other hand, while nowadays Photogrammetry has
become a widely accessible and popular technique, the
achievement of adequate results in the final products is not
trivial. If the desired output is a good-looking 3D model for
mere visualization purpose, also a low-quality point cloud can
be textured to produce a model, whose geometric content is not
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W11, 2019
GEORES 2019 – 2nd International Conference of Geomatics and Restoration, 8–10 May 2019, Milan, Italy
This contribution has been peer-reviewed.
https://doi.org/10.5194/isprs-archives-XLII-2-W11-1165-2019 | © Authors 2019. CC BY 4.0 License.
1165

sufficient for deeper analysis or to plan restoration actions. For
such a reason, a set of guidelines may help non-expert users to
improve their approach to SfM. On the other hand, to
completely control the photogrammetric process a theoretical
and practical background are both required, which should
include several components: digital camera technology, network
geometry, image processing, adjustment theory, elementary
surveying, GNSS, photographic technique, photogrammetric
and computer vision principles, and so on. But the knowledge
of all these factors is still not enough in the case of poor
experience. Training is then a pivotal task to achieve a qualified
skill to accomplish SfM (Rutzinger et al., 2018). This is
particularly important when dealing with CH, which may
require a level of accuracy and resolution of the final products
that is bigger than the ones needed in other domains, such as in
the Geosciences (Eltner et al., 2016).
In 1994, Waldhäusl and Ogleby published a paper reporting
what it is well known as the “3x3 Rules” for photogrammetry
with non-metric cameras in the field of CH documentation.
Such a set of guidelines were set up first to help students to
carry out good photogrammetric projects. Then they have been
extended and submitted to the CIPA-committee to become a
standard. Following that concept, this paper would like to
suggest an up-to-date version of the “3x3 Rules” to be used
within modern SfM for surveying projects in the field of CH. In
particular, the new proposed guidelines (Sect. 2) have been
thought with the aim of supporting people to learn how to
accomplish photogrammetric projects by themselves, without
the help of experts. This capability could be useful to operate in
those remote archeological areas that are difficult or even
impossible to be reached by external experts, as discussed at the
beginning of this section. Of course, as the authors of the “3x3
Rules” in 1994 did make a proposal to be integrated and
discussed in the scientific and practitioner community, the
humble intention of this paper is again to make some
suggestions only and to open a debate.
In order to better understand how the new guidelines for SfM
may help in real applications, we aimed at the 3D reconstruction
of Ziggurat Chogha Zanbil in Iran (Sect. 3). Though this can be
accessed without any problem, the place has some
characteristics that are typical of those remote archaeological
areas where training local people would be a more viable way to
collect 3D data. After the presentation of this example, some
comparisons are proposed and discussed in Section 4.
Eventually some conclusions are drawn in Section 5.
2. GUIDELINES FOR SFM PHOTOGRAMMETRY
2.1 Review of the “3x3 Rules”
The “3x3 Rules” were organized in three main sections, each of
them consisting of three subsections. The first section deals
with aspects related to the photogrammetric network geometry,
such as the control information (scale bars, plumb-lines) and
camera station geometry. It deserves to be observed that in the
network two main functions are distinguished: the necessity of
linking photos covering the whole object to survey and the
presence of stereopairs for stereoplotting, that at the time was
the approach used to derive 3D outputs. The same is
recommended for the production of orthophotos and
rectifications, which should be based on photos parallel to the
main facades. The second section entails the photographic
rules, including camera and lens selection, setup, and
illumination. The third section lists some organization rules: the
preparation of sketches, protocols and the final check.
Of course, some of these rules are now obsolete due to the
transition from analogue to digital imaging technology.
However, several practical rules are still valid, while others
need to be updated to account for the acquisition methodology
necessary when using SfM. In addition, the “3x3 Rules”
concern the data acquisition phase, which is only the first step
of the photogrammetric process. Other rules may be added to
guide the successive processing phase: camera calibration and
image orientation, dense surface matching, point cloud
processing, quality assessment, and production of final outputs.
On the other hand, data acquisition is the crucial stage in order
to set up a solid photogrammetric project, also because this task
cannot be assisted as it happens in some popular
photogrammetric software packages, where the user is guided
along with the processing workflow. Furthermore, processing
may be also done by expert people in a remote laboratory.
2.2 Guidelines for data acquisition with SfM
The conclusion of the analysis reported in the previous
subsection is the relevance to have some guidelines to support
data acquisition when using SfM in archeological applications.
Following the scheme of the “3x3 Rules”, we discuss in the
following geometric, imaging and organizational aspects. We
consider in this section the planning of data acquisition from
ground-based camera stations. Since drones (Granshaw, 2018a)
are widely used in modern photogrammetric projects, when
allowed by local regulations, the readers are suggested to refer
to the specific literature (O’Connor et al., 2017; Pepe et al.,
2018).
2.3 Geometric aspects
2.3.1 Control information. Very often, the precise
georeferencing of a single project in a mapping reference
system is not necessary or may be done using navigation-grade
GNSS sensors, data from smartphones or using online
geoportals. A local reference system is generally sufficient.
Then the control information for the photogrammetric project
only requires defining the scale of the 3D model and the local
plumb-line direction. The following hints could be stated about
this point:
- some known distances (e.g., scale bars) on the object may
suffice to define the scale, provided that:
- distances are comparable to the object’s size (do not use a
one-metre bar to fix the distance for an object sizing 100
m!);
- distances are taken in different orthogonal directions,
especially if the object has a complex shape;
- endpoints of each distance should be well defined; if
possible, use targets to this purpose;
- define one or more plumb-lines to set up the vertical
direction; and
- if the object has a complex shape or spans over a large
area, split the survey into more photogrammetric projects,
to be joined using some ground control points (GCP)
measured using a theodolite. Some rules about the
number of GCPs to be used may found in Scaioni et al.
(2018).
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W11, 2019
GEORES 2019 – 2nd International Conference of Geomatics and Restoration, 8–10 May 2019, Milan, Italy
This contribution has been peer-reviewed.
https://doi.org/10.5194/isprs-archives-XLII-2-W11-1165-2019 | © Authors 2019. CC BY 4.0 License.
1166

2.3.2 Network geometry. The geometry of the
photogrammetric network is defined by the 3D position and
attitude of each camera station. When using SfM, two different
aspects should be balanced: (1) convergent photos and long
baselines (i.e., the distances between camera stations) help the
stability of the block geometry; (2) the presence of small (less
than 10°) angles between adjacent photos makes easier the
image matching at both orientation and dense matching stages
(Barazzetti et al., 2009; 2011). Stereoplotting is generally not
use any more for 3D modelling, which is based on extracting
information from the point cloud obtained from dense
matching. Considering these points, the following rules may be
remembered:
- fix the average and the minimum value for the photo scale
depending on the design resolution and precision. As a
rule-of-thumb, consider the ground sampling distance
(GSD), i.e., the average size ot the pixel footprint on the
object as reference value. Remember that:
GSD = d (pz / c) (1)
While the pixel size (pz) and the focal length (c) both
depend on the adopted sensor, the average distance
camera-object (d) can be selected, provided that geometric
constraints in the nearby may limit the positioning of
camera stations;
- the image acquisition should be based on sequences with
80% overlap and small relative rotation angles between
consecutive images;
- sequences should be organized to follow the shape of the
object along lines or rings;
- in the case of linear sequences, include some convergent
photos (“arch bridge” rule), which play a twofold function
of strengthening the network geometry (see Fraser, 1996)
as well as to improve the visibility of those surfaces that
are not parallel to the main sequence. An example of such
a sequence in depicted in Figure 1;
- Add some 90° rolled photos to improve camera
calibration (roughly, one rolled photos every 10-15
photos may be enough);
- add sub-blocks to reconstruct details such as doors,
decorations, bas-reliefs, regions with occlusions;
- capture images from half the object’s height; if necessary,
organize two overlapping sequences (keep at least 60%
sidelap between them);
- each item related to control information (plumb-lines,
scale bars, targets) should be captured in at least three
convergent images;
- check multiple coverage;
- add photos parallel to the object’s facades to produce
orthophotos or rectifications; and
- when using targets or scale bars/plumb-lines, take also
photos from the same positions after removing them.
While all photos will be processed together for image
orientation, only the ones without targets (or scale
bars/plumb-lines) will be exploited for surface
reconstruction or texturing.
Figure 1. Example of acquisition of a linear sequence of images
including alternate convergent photos (“arch bridge”
rule).
2.4 Imaging aspects
2.4.1 Camera selection. Today the panorama of the available
sensors for photogrammetric data acquisition is quite huge,
including frame and panoramic cameras (Barazzetti et al.,
2018). Cameras embedded in smartphones and action cameras
have been proved to work well for SfM Photogrammetry as
well. For other camera models or special lenses (e.g., fish-eyes)
the reader is recommended to look the specialized literature. In
the case of consolidated frame-camera technology, the
following recommendations should be paid attention:
- try to avoid long focal lens (longer than 50 mm equivalent
lens for 24x36 mm full-format), except than in specific
projects requiring acquisition from large distances;
- in general, large-format sensors may provide better image
quality and less noisy images. Please note that here we
refer to the physical sensor size, not to the number of
pixels; and
- the use of more cameras is not suggested; if this option is
needed, for example because of merging ground-based
and drone-based photos, organize independent projects to
be merged afterwards using GCPs, manually or
automatically selected common points, or by merging
point clouds (see Scaioni et al. 2018).
2.4.2 Camera setup. Some of the rules proposed in
Waldhaeusl and Ogleby (1994) are still valid, to be integrated
by additional requirements of digital imaging technology:
- do not change focal lens during your project:
- turn off any autofocusing option;
- fix focal lens in the case of zoom-lens (use preferably the
end position or fix the focal lens using a tape);
- turn off any function which may modify the original
image geometry, such as spotting, automatic rotation of
portrait photos, denoising filters;
- check out the correct recording of EXIF info in the image
files;
- use the largest image size format available in the camera;
and
- do not use small compression rates (<95% in the case of
JPG).
2.4.3 Scene illumination. In digital imaging the problem of
poor lighting cannot be overcome by rising the sensibility
(higher ISO values), which result in more noise in the image
content. Some recommendations should be remembered:
- select the best time of day, to guarantee a sharp lighting
and mitigate the effect of shadows;
- do not operate in windy conditions, that also change
shadows quickly;
- use tripod or other stabilizing tools; and
- shot photos using timer function in the case of hand-held
acquisition.
2.5 Organizational aspects
Under this aspect, the content of the “3x3 Rules” is still valid
and should be carefully considered. For this reason, we do not
revise this topic here.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W11, 2019
GEORES 2019 – 2nd International Conference of Geomatics and Restoration, 8–10 May 2019, Milan, Italy
This contribution has been peer-reviewed.
https://doi.org/10.5194/isprs-archives-XLII-2-W11-1165-2019 | © Authors 2019. CC BY 4.0 License.
1167

3. APPLICATION
3.1 The Ziggurat Chogha Zanbil
The southwest of Iran - an area not far from the domains of the
Zagros Mountains, between two large rivers named Karoon and
Dez - is the birthplace of the great kingdom of Elam in ca. 4000
B.C. (Potts, 1999). As can be seen in Figure 2, this region is
located 90 km north of the city of Ahvaz and 35 km south of the
ancient city of Susa (Emami, 2012).
Figure 2. Geographic location of the Ziggurat Chogha Zanbil
(Iran).
Ziggurat Chogha Zanbil (Carter, 1996) is located in a settlement
founded by the Elamite king Untaš Napiriša (The Elamite name
of this structure is Ziggurat Dūr Untash), see Figure 3. The
outer enclosure wall is about 1,300 m x 1,000 m while the
second and third inner enclosures size 400 m x 400 m and 200
m x 180 m, respectively (see a map in Fig. 4). The remains of
the Ziggurat stand up to a height of more than 25 m, structured
on three levels above the surrounding pavement (see Figs. 3 and
4). Originally, it consisted of five levels rising up to 53 m. The
material used for construction is mud-bricks for the core, which
are covered of baked bricks layer 2 m thick.
Figure 3. Aerial view of the Ziggurat Chogha Zanbil.
Figure 4. Topographic map of Ziggurat Chogha Zanbil (from
Ghirshman, 1966).
3.2 Photogrammetric data sets
The surveyed area of the Ziggurat Chogha Zanbil was inside the
3rd enclosure wall, chiefly consisting of the Ziggurat building
(approx. 20,000 m2). Because of the construction’s complexity,
four photogrammetric data sets were collected:
1. Ground-based (GB) photos;
2. Low-angle oblique UAV photos (LAOUAV);
3. High-angle oblique UAV photos (HAOUAV); and
4. Nadir UAV photos (NUAV).
Figure 5 reports some typical camera standpoints for different
blocks, which are described in the following paragraphs.
Figure 5. Typical camera standpoints for the photogrammetric
data sets.
In addition, seven GCPs were measured. These were positioned
either in the surrounding area of the Ziggurat and on the
Ziggurat itself. The CGP positions are shown in Figure 6. Some
GCPs were located off-ground in order not let them all lie on a
plane. Indeed, GCP A6 is located on the first level of the
Ziggurat and GCP A7 is on the top of the upper inner part. The
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W11, 2019
GEORES 2019 – 2nd International Conference of Geomatics and Restoration, 8–10 May 2019, Milan, Italy
This contribution has been peer-reviewed.
https://doi.org/10.5194/isprs-archives-XLII-2-W11-1165-2019 | © Authors 2019. CC BY 4.0 License.
1168

GCP coordinates were measured using a multi-frequency GNSS
sensor (SOUTH – Galaxy G1 Plus).
Figure 6. - Planimetric positions of GNSS GCPs.
3.2.1 Ground-based block (GB). The GB Data Set (Fig. 7)
was planned to cover the vertical facades of the Ziggurat from
camera stations located at approximately 1.7 m from ground. A
Sony alpha 7RII SLR (single-lens reflex) camera equipped with
16 mm lens was used (see Table 1). It consisted of four linear
sequences including some convergent photos, as suggested in
paragraph 2.3.2. It was ensured that terrestrial photos were
taken at a sufficient distance (~15-20 m) with 80% overlap to
guarantee a GSD between 2-3 cm. Some convergent photos
around corners were collected to connect linear sequences.
Obviously, this data set is not able to capture the upper part of
the Ziggurat.
Figure 7. Camera poses of GB Data Set.
Table 1. Main properties of sensors adopted to collect
photogrammetric data sets.
3.2.2 Oblique UAV blocks (LAOUAV/HAOUAV). A couple
of UAV data sets using oblique setup were thought as a trade-
off to cover both vertical walls and the upper part of the site.
For this reason, two different inclination angles have been tried
(30°- 60°), each of them consisting on a circular sequence
around the Ziggurat. Dense image matching of oblique images
permits to include the façade description and the building
footprints in the models. All UAV missions were operated using
a DJI Phantom 4 Pro Plus carrying an 8.8 mm lens camera (see
Table 1).
In the case of low-angle oblique UAV block (LAOUAV), one
full circular (average diameter d=140 m) image sequence was
captured at approximately 20 m relative height from ground and
orientation about 30q from the local horizontal plane. This
block resulted in 10-12 images per facade.
In the case of high-angle oblique UAV block (HAOUAV), to
avoid blurry images caused by wind exist on surveying site, two
full circulars (average diameter d1=210 m and d1=270 m)
sequences at approximately 45 m and 50 m relative height from
ground with orientation about 60q, while the onboard camera
was roughly 45q oriented toward the Ziggurat (see Fig. 8). The
number of images for each facade (15-17) was higher than in
the case of LAOUAV Data set.
Figure 8. Camera poses of HAOUAV Data Set, which consists
of two circular sequences around the Ziggurat.
3.2.3 Nadir UAV blocks (NUAV). The UAV block based on
nadir photos (see Fig. 9) is suitable for collecting the ground
surface and the upper part of the construction but cannot
properly cover vertical surfaces. UAV surveys are usually nadir,
which means that the images are shot with the camera axis
along the vertical direction; they provide both a forward overlap
between shots and a side one between strips, allowing the
reconstruction of the surveyed t object in 3D (Vacca et al.,
2017). In our case study, an approximate 50 m relative height
from ground on a linear grid pattern was selected. A total
number of 160 images were taken in order to cover all 3rd
enclosure wall area (approx. 4,000 m2).
Figure 9. Camera poses of NUAV Data Set.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W11, 2019
GEORES 2019 – 2nd International Conference of Geomatics and Restoration, 8–10 May 2019, Milan, Italy
This contribution has been peer-reviewed.
https://doi.org/10.5194/isprs-archives-XLII-2-W11-1165-2019 | © Authors 2019. CC BY 4.0 License.
1169

3.3 Data processing
3.3.1 Pre-processing. After the acquisition stage one should
manually verify the quality of the obtained images. Where
photos with shifted focus location or totally out of focus should
be removed, the latter is also valid for images blurred during the
capture process (e.g., tremble during hand-held acquisition in
low-light conditions – see recommendations at par. 2.4.3). This
check-out is suggested to be done on-site, so that some photos
may be recaptured, if necessary. In general, the acquisition of a
good data set will ensure the quality of next modelling stages,
as well it will save time and financial costs, especially in areas
not easy to access.
In some cases, it is hard for one to choose the best illumination
conditions. Therefore, some processing of the images could be
applied, just for the purpose of extracting useful information
from them (e.g., in case of large shadows). Where most of the
suggested actions are related to colour balancing, exposure
equalization and denoising may be applied (Ballabeni et al.,
2015). While some photometric manipulations do not affect the
following reconstruction, other manipulations such as cropping,
resizing or rotating the images are not suggested. During the
pre-processing actions, it is important that the EXIF information
is not lost, since its valuable information of the sensor size and
focal length parameters is crucial for the camera calibration. It
is worth noting, that all pre-processing manipulations should be
applied to the whole dataset.
3.3.2 Structure-from-Motion Photogrammetry. Nowadays,
there is a great variety of photogrammetric software solutions
implementing the SfM processing pipeline. Here we adopted
Agisoft Metashape® (AMs) ver. 1.5.0, which is a popular SW
package adopted in several domains. The same processing
pipeline was adopted for different blocks, which were processed
independently.
The image orientation (“alignment” in the AMs jargon) was
operated by using images at original full resolution (while AMs
also allows to work with subsampled images in the case of very
large projects or when a lower resolution of the outputs is
enough). The camera parameters’ estimation was performed
during the bundle adjustment applied to compute camera
orientation and 3D coordinates of automatically extracted tie
points (Barazzetti et al., 2011), which define the so-called
“sparse point cloud.” Camera calibration and orientation were
computed per each data set of photos.
After “alignment”, the dense matching function was applied to
densify the “sparse point clouds” and obtain “dense point
clouds.” Further edit of the point cloud is suggested, where all
of points created outside the area of interest should be removed.
After obtaining dense point clouds (see, e.g., Figs. 10 and 11)
from different data sets, one can combine them in one single
block, which will improve the overall quality and increase the
details in parts where single blocks could not provide a
complete point cloud. For example, from NUAV Data Set we
obtained a point cloud which did not contain detailed
information of facades. On the contrary, GB Data Set resulted in
opposite performances. A combination between point clouds
achieved from different blocks may performed better (see an
example in Fig. 12). The merge of multiple point clouds can be
done using common GCPs. When this solution cannot be
pursued (e.g., GB Data Set does not contain GCPs shared with
UAV blocks), a manual measurement of corresponding features
is suggested to merged point clouds.
Figure 10. Top view of the LAOUAV dense point cloud.
Figure 11. Top view of the NUAV dense point cloud.
Figure 12. Top view of the merged point cloud.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W11, 2019
GEORES 2019 – 2nd International Conference of Geomatics and Restoration, 8–10 May 2019, Milan, Italy
This contribution has been peer-reviewed.
https://doi.org/10.5194/isprs-archives-XLII-2-W11-1165-2019 | © Authors 2019. CC BY 4.0 License.
1170

4. DISCUSSION
The data processing pipeline described in Subsection 3.3 was
independently applied to all data sets captured on the Ziggurat
(see Subsect. 2.3). In Table 2 some details about obtained point
clouds and their accuracy can be found. It can be noted that the
highest accuracy was achieved from NUAV Data Set. On the
contrary, there is no information about the accuracy of GB Data
Set since no GCPs were present. On the other hand, each point
cloud contributed to a merged point cloud with a global
accuracy of approximately 8 cm in term of RMSE (Root Mean
Square Error) on GCP residuals. In the meantime, the latter
effect is clear visible on the facades of the temple, where the
merged clouds yield higher level of details and lack of holes
with missing information. Nevertheless, a more thorough
comparison between the clouds is needed to determine the co-
registration accuracy between the individual blocks. For that
purpose, the tool Cloud-2-Cloud (C2C) distance in the open-
source software package CloudCompare® (www.
cloudcompare.org) is suggested. In the comparison phase it is
preferred to use the cloud that is the most accurately
georeferenced as reference and compare others to that. In our
case, NUAV point cloud was selected to this purpose. It should
be noted, that point clouds were subsampled at 10 cm minimum
distance between points in order to save memory and
computational time.
Point cloud Data Set # points
[M]
RMSE of GCP
resi
duals [cm]
GB 75.0 Not available
LAOUAV 141.6 19.8
HAOUAV 67.4 9.6
NUAV 80.2 3.4
Merged 364.2 8.1
Table 2. Point clouds’ information.
Figure 13. C2C distance comparison between NUAV and HAOUAV clouds.
Figure 14. Facade view of the GB cloud with an aspect to the details of the cloud.
Figure 15. Facade view of the NUAV cloud representing different level of details.
Figure 16. Facade view of the merged point cloud.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W11, 2019
GEORES 2019 – 2nd International Conference of Geomatics and Restoration, 8–10 May 2019, Milan, Italy
This contribution has been peer-reviewed.
https://doi.org/10.5194/isprs-archives-XLII-2-W11-1165-2019 | © Authors 2019. CC BY 4.0 License.
1171

In Figure 13 an example of comparison (NUAV and HAOUAV
point clouds) is shown. One can note that the actual differences
between both clouds are quite small (<10 cm in absolute value),
and comparable with the final point cloud resolution. The
largest dissimilarities are in areas affected by strong shadows, or
in where less GCPs were available for the merging point clouds.
In addition, it is apparent the effect on the oblique clouds and
their contribution to the details of the facades. Of course, the
integration of point clouds from GB and NUAV Data Sets
represents the trade-off between those two acquisition modes
(Figs. 14 and 15), but the overall contribution of all Data Sets to
the final cloud is clear in Figure 16.
4. CONCLUSIONS
In this paper some guidelines to accomplish photogrammetric
data acquisition of an archeological site have been presented.
The suggested methodology has been drawn in view of the
application of Structure-from-Motion Photogrammetry (SfM).
In particular, these guidelines have been defined to allow non-
expert people to operate in remote areas where archeological
sites may be located.
The presentation of a case study related to the Ziggurat Chogha
Zanbil (Iran) has demonstrated that the proposed guidelines are
useful to drive people who have to plan and operate
photogrammetric data acquisition in a typical remote
archeological area.
Those guidelines do not have a definitory character, but they
would call for the attention of the scientific community towards
the necessity to develop and share best practices and standards.
Here we have limited the attention to the data acquisition stage,
that necessarily has to be done on site. For this reason, training
of local people is a fundamental task. On the other hand, also
the successive steps of the SfM pipeline, such as the point cloud
generation, modelling and the extraction of final outputs, need
to be paid attention and to be focused in future papers.
Acknowledgements
The authors would like to thank Agisoft company (St.
Petersburg, Russia) for the availability of trial licenses of
Agisoft Metashape®.
REFERENCES
References from books
Carter, E., 1996. ČOḠĀ ZANBĪL. Encyclopædia Iranica, Vol. VI, Part
1, pp. 9-13, available online at http://www.
iranicaonline.org/articles/coga-zanbil (last accessed on 21st March
2019).
Fraser, C.S., 1996. Network design. Close Range Photogrammetry and
Machine Vision. (Ed. K.B. Atkinson). Whittles Publishing, Caithness,
Scotland. 371 pages: 256-281.
Ghirshman,R., 1966. Tchoga Zanbil(Dur-Untash) I. La ziggurat.
MDAFI 39, Paris, 1966.
Hartley, R., Zisserman, A., 2006. Multiple View Geometry in computer
vision. Cambridge University Press, UK.
Luhmann, T., Robson, S., Kyle, S., Boehm, J., 2014. Close Range
Photogrammetry: 3D Imaging Techniques – 2nd Edition . Walter De
Gruyter Inc., Germany, 684 pages.
Potts, D.T., 1999. The Archaeology of Elam Formation and
Transformation of an Ancient Iranian State From. Cambridge World
Archaeology.
References from journals
Barazzetti, L., Binda, M.L., Scaioni, M., Taranto, P., 2011.
Photogrammetric survey of complex geometries with low-cost software:
application to the ‘G1’ Temple in Myson, Vietnam. J.Cult. Heritage,
12(2011): 253-262.
Eltner, A., Kaiser, A., Castillo, C., Rock, G., Neugirg, F., Abellán, A.,
2016. Image-based surface reconstruction in geomorphometry–merits,
limits and developments. Earth Surf Dyn, Vol. 4:359-389.
Emami, M., Trettin, R., 2012. Mineralogical and chemical
investigations on the ceramic technology in Čogā Zanbil, (Iran, 1250
B.C.). Period Mineral., vol. 81(3):359–377
Ghirshman, R., 1960. The Ziggurat of Tchoga-Zanbil. Sci. Am., Vol.
204(1): 68–76
Granshaw, S.I., 2018a. RPV, UAV, UAS, RPAS … or just drone?
Photogramm. Rec., 32(162): 160-170.
Granshaw, S.I., 2018b. Structure from Motion: Origins and Originality.
Photogramm. Rec., 33(161): 6-10.
Luhmann, T., Fraser, C.S., Maas, H.-G., 2016. Sensor Modelling and
Camera Calibration for Close Range Photogrammetry. ISPRS J.
Photogramm. Remote Sens., Vol. 115: 37-46.
O’Connor, J., Smith, M.J., James, M.R., 2017. Cameras and settings for
aerial surveys in the geosciences: Optimising image data. Prog Phys
Geog, Vol. 41:325-344.
Pepe, M., Fregonese, L., Scaioni, M., 2018. Planning Airborne
Photogrammetry and Remote-Sensing Missions with Modern Platforms
and Sensors. Eur. J. Remote Sens., Vol. 51(1): 412-435, DOI:
10.1080/22797254.2018.1444945.
Snavely, N., Seitz, S.M., Szeliski, R., 2006. Photo tourism: Exploring
photo collections in 3D. ACM Trans. Graphics, Vol. 25(3): 835-846.
Ullman, S., 1979. The interpretation of structure from motion. Proc.
Royal Soc. London, Series B, Biolog. Sci., Vol. 203:405-426.
Vacca, G., Dessì, A., Sacco, A., 2017. The Use of Nadir and Oblique
UAV Images for Building Knowledge. ISPRS Int. J. Geo-Inf., Vol.
6(12), paper no. 393.
References from proceedings
Ballabeni, A., Apollonio, F. I., Gaiani, M., Remondino, F., 2015.
Advances in image pre-processing to improve automated 3D
reconstruction. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.,
Vol. XL-5/W4, pp. 315-323.
Barazzetti, L., Remondino, F., Scaioni, M., 2009. Combined use of
photogrammetric and computer vision techniques for fully automated
and accurate 3D modeling of terrestrial objects. In: Videometrics,
Range Imaging, and Applications X, Proc. of SPIE, paper no. 74470M.
Barazzetti, L., Forlani, G., Remondino, F., Roncella, R., Scaioni, M.,
2011. Experiences and achievements in automated image sequence
orientation for close-range photogrammetric projects. In: Videometrics,
Range Imaging, and Applications XI, Proc. of SPIE, paper no. 80850F.
Barazzetti, L., Previtali, M., and Roncoroni, F., 2018. Can we use low-
cost 360 degree cameras to create accurate 3D models? Int. Arch.
Photogramm. Remote Sens. Spatial Inf. Sci., Vol. XLII, Part 2:69-75.
Rutzinger, M., Bremer, M., Höfle , B., Hämmerle, M., Lindenbergh, R.,
Oude Elberink, S., Pirotti, F., Scaioni, M., Wujanz, D., Zieher, T.,
2018. Training in Innovative Technologies for Close-Range Sensing in
Alpine Terrain.ISPRS Ann. Photogramm. Remote Sens. Spatial Inf.
Sci., Vol. IV, Part 2, pp. 239-246, DOI: 10.5194/isprs-annals-IV-2-239-
2018.
Scaioni, M., Crippa, J., Corti, M., Barazzetti, L., Fugazza, D., Azzoni,
R., Cernuschi, M., Diolaiuti, G.A., 2018. Technical Aspects Related to
the Application of SfM Photogrammetry in High Mountain. Int. Arch.
Photogramm. Remote Sens. Spatial Inf. Sci., Vol. XLII, Part 2: 1029-
1036.
Waldhaeusl, P., Ogleby, C., 1994. 3 x 3 rule for simple
photogrammetric documentation of architecture. Int. Arch
Photogramm. Remote Sens. Spat. Inf. Sci., Vol. XXX, Part 5: 426-429.
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W11, 2019
GEORES 2019 – 2nd International Conference of Geomatics and Restoration, 8–10 May 2019, Milan, Italy
This contribution has been peer-reviewed.
https://doi.org/10.5194/isprs-archives-XLII-2-W11-1165-2019 | © Authors 2019. CC BY 4.0 License.
1172