You are on page 1of 9

Journal of Archaeological Science 46 (2014) 173e181

Contents lists available at ScienceDirect

Journal of Archaeological Science


journal homepage: http://www.elsevier.com/locate/jas

A comparative assessment of structure from motion methods for


archaeological research
Susie Green a, *, Andrew Bevan a, Michael Shapland b
a
UCL Institute of Archaeology, 31-34 Gordon Square, London WC1H 0PY, UK
b
Archaeology South-East/UCL Centre for Applied Archaeology, 2 Chapel Place, Portslade, East Sussex BN41 1DR, UK

a r t i c l e i n f o a b s t r a c t

Article history: This paper addresses the use of open source, structure from motion methods for creating 3d pointclouds
Received 17 August 2013 from photographs and compares these with alternative workflows in other software, and relative ac-
Received in revised form curacy compared to other 3D modelling methods. It describes a series of case studies that use structure
17 January 2014
from motion to record standing buildings and create digital elevation models. Looking at other recording
Accepted 21 February 2014
Available online 20 March 2014
techniques it finds that structure from motion can produce better results than traditional techniques
such as plan drawing, topographic survey and photogrammetry, and is cheaper and more accessible than
new techniques such as laser scanning and LiDAR, although it is less accurate in some regards. It
Keywords:
Structure from motion
demonstrates that good accuracy can be achieved if careful measurements are made, and concludes that
Computer vision it has great potential for widespread archaeological application.
Bundler Ó 2014 Elsevier Ltd. All rights reserved.
Open source
Photogrammetry
Multi-view stereo

1. Introduction et al., 2012). Photoscan enables georeferencing and creation of


textured models from the resulting pointcloud and creation of
Structure from motion (SfM) is the name given to the extraction digital elevation models. However at a few thousand US dollars at
of three dimensional data and camera positions from a collection of current prices for the professional edition (Agisoft, 2013) it is
photographs. It is the computer vision equivalent of a human’s beyond the budget of many archaeologists, and there is nothing
ability to understand the 3d structure of a scene as they move that can be achieved using this software that cannot be reproduced
through it (Szeliski, 2010). In practical terms this is a photogram- using open source programs. Indeed, open source technology is not
metric approach that involves taking a collection of photographs of only important to archaeology because it is free to use, but more
an object or scene that we wish to record, and processing them importantly because the source code is free to anyone to interro-
through a series of computer programs to create a 3d pointcloud gate and further develop. If it is embraced by archaeologists at an
representing the surfaces of the objects in the photographs. The early stage then it can be developed in a direction that is advan-
resulting pointcloud can then be processed further to create, tageous to archaeology’s specific research and management
among other things, 3d models, digital elevation maps, rectified agendas. In addition, open source solutions allow the archaeologist
images, scaled plans, elevations and cross sections. This paper in- to retain ownership of the data throughout the process. Archaeo-
vestigates the use of the open source programs Bundler and Patch logical records are often paid for by public money, and it is there-
Based Multi-View Stereo (PMVS) to automatically create point- fore important that they belong to the public and can be accessed
clouds without the need for human input. by them, or at least accessed on their behalf. Commercial software
Bundler and PMVS2 are both released under the GNU public companies often store data in formats or under licensing arrange-
license. Most papers describing the use of SfM in archaeology have ments that mean they are not available to the public, with this
focussed on the use of Agisoft’s Photoscan which is the commercial being a well-known issue for laser-scanning archives. For various
software that is most directly comparable to Bundler and PMVS2 reasons the owners of the data (the public) may lose access to the
(De Reu et al., 2013; Olson et al., 2013; Verhoeven, 2011; Verhoeven software with which to read it; for example the license may no
longer be affordable to the organisation holding the data, the
company producing the software may go bust, or the software may
* Corresponding author. Tel.: þ44 (0)7738242888.
E-mail addresses: susan.green.10@ucl.ac.uk, susiezgreen@gmail.com (S. Green). simply no longer be sold. In the case of open source software the

http://dx.doi.org/10.1016/j.jas.2014.02.030
0305-4403/Ó 2014 Elsevier Ltd. All rights reserved.
174 S. Green et al. / Journal of Archaeological Science 46 (2014) 173e181

code with which stored data is encoded is freely available to 2.3. SfM algorithms and software considerations
anyone, so if the software needed to read it is no longer available, it
can always be rewritten. In traditional photogrammetry, either the positions of the
The SfM open source toolchain needed to produce a textured cameras, or the position of some points that are visible in more than
model or digital elevation map consists of several different pieces of one image, are known. These are triangulated in order to find the
software and this multi-package workflow can be daunting for location of other points in the photographs. In Structure from
users when compared to the end-to-end solution offered by com- Motion, matches are made between many points across many
mercial software such as Photoscan. However this modular images without prior knowledge of the camera position (Lowe,
approach also has its advantages as it allows the archaeologist to 2004). The process involves two steps:
understand each step in the process, to see where problems lie, and
vary his/her pipeline to get the best results from each dataset. The  step 1. The photographs are examined to find matching points
adoption of Bundler and PMVS2 by ArcheOS, the open source across several images and these are used to calculate the posi-
operating system for archaeologists (ArcheOS, 2011), is an tion of the cameras relative to each other.
encouraging start. Ducke and colleagues have already documented  step 2. Now that the camera positions are known, the location of
the recording of an excavation of a burial site using Bundler and many points can be plotted in space, giving a dense recon-
PMVS2, but their discussion focused on the robustness, simplicity struction of the shape of the objects that were photographed.
and affordability of the process, rather than considering the accu-
racy of the results (Ducke et al., 2011). This paper aims to build on The first step is iterative, with the results being refined as more
that work by directly comparing SfM with other recording tech- points are added. As each camera is added a check is made that the
niques used by archaeologists in order to assess the accuracy (the matches fall within the expected parameters, and if not it is
degree to which the results match their real world values) and rejected. As the number of images increase, the time taken to assess
precision (the resolution of the results). each camera becomes exponentially longer which means that for
very large image sets the process is impractically slow (Snavely
et al., 2006). The second step of the process uses the same princi-
2. Methods
ples as traditional photogrammetry, but is capable of processing
huge numbers of points. The result is a pointcloud that looks similar
2.1. The investigation
to those produced by a laser scanner.
For the case studies in this investigation the first step was car-
The possible applications of SfM in archaeological research are
ried out using Bundler and the second using PMVS2. A major
numerous. This paper uses four case studies to illustrate some of
drawback of the PMVS2 dense pointcloud generation is that its
the areas in which it could be of benefit to archaeology, and com-
processing time scales badly. This problem is addressed by a pro-
pares the results with more traditional recording techniques. The
cess known as CMVS, which breaks the task up into smaller chunks
first case study looks at the recording of standing buildings
(Furukawa et al., 2010). Currently, the above pipeline can be made
compared to traditional plan drawing. The second demonstrates
simpler still via VisualSfM, which also has a further advantage of
how SfM can be used to record highly irregular shapes and create
using multiple computer cores to speed up the process of locating
cross sections. The next case study looks at the use of SfM to create
camera positions and parameters (Wu et al., 2011; not open source,
digital elevation models. In order to cover large areas of ground a
but is free for non-commercial use), and includes a graphical user
kite was used to elevate the camera into the air. A surface was
interface that makes implementation simpler. Another solution
created from the pointcloud that can be represented by x and y
would be the use of ArcheOS, the open source operating system for
geographical coordinates, holding the z (elevation) information as
archaeologists (ArcheOS, 2011), which incorporates the bundler
an attribute. These models were compared with a topographic
pipeline and simplifies it with a python wrapper. Other solutions
survey map created using total station survey data, and with a
that are free for non-commercial use include the web based Pho-
digital terrain model created from LiDAR. The final case study uses a
tosynth from Microsoft, which only produces a low resolution
small scale digital elevation model to examine surface deformation.
pointcloud but it is very fast and is therefore useful for testing sets
In each of the case studies a number of control points were sur-
of photographs (Microsoft, 2014), and 123D Catch from Autodesk,
veyed. The accuracy of an SfM approach was established using the
another fast, free, web based solution which can also produce
distance between the real world points and their equivalents in the
models and surface textures, but the results are mixed and are
pointcloud. This paper also examines the precision achieved by
better suited to discrete objects rather than large features or
SfM, which can be determined by the amount of detail that is
landscape. It is designed as part of a 3D printing pipeline (Autodesk,
evident in the pointcloud, and looks at the relationship between
2014).
accuracy and precision and its implications when interpreting
Bundler and most other structure from motion methods of this
results.
kind have no concept of the absolute world space represented in
the photographs so, instead, pointclouds are created in an arbitrary
2.2. Source photographs space and must be georeferenced in a further step. To do this we use
distinctive markers or features within our photographic subjects,
In order to create a pointcloud without gaps it is usually and survey them to find their real-world locations. These points can
necessary to use a large number of images. These case studies use be picked out of the pointcloud and a comparison made between
between 250 and 650 photographs each. Each point in the point- their geographical coordinates and their virtual coordinates, and
cloud must be visible in at least three photographs taken from the difference can be used to calculate an affine transformation to
different positions. The number of images necessary is dependent georeference the pointcloud correctly (e.g. using CloudCompare;
on the complexity of the shape of the subject. The angles between Girardeau-Montaut, 2014). Further manipulation of pointclouds
each pair of photographs should not be more than 15 degrees, but can be performed using the open source program Meshlab
also should not be too small or PMVS2 will be unable to correctly (Meshlab, 2013), which will also allow the creation of polygon
locate the points in space. All the case studies described here use a meshes and the texturing of surfaces using the original photo-
low cost digital camera (Canon G11). graphs. It should be noted that this entire pipeline is possible
S. Green et al. / Journal of Archaeological Science 46 (2014) 173e181 175

within the commercial program Photoscan. Table 1 contains an SfM model is a closer match to the points than the drawing as can
outline of each part of the process for the software described above. be seen in the top section around the circular windows. The
drawing was created from a photograph that was rectified using the
3. Case studies: standing buildings same total station points (Shapland, 2012), and this fairly approx-
imate polynomial transformation of the image is likely to have
3.1. St Andrew’s Church, Jevington introduced some distortion. Structure from Motion appears to
create pointclouds that are free of perspective distortion.
The first case study looks at the potential of structure from
motion to record plans and elevations of standing buildings as a 3.2. St Benedict’s Basilica at Bury St Edmunds
replacement for traditional scale drawing. The varied and non-
reflective surfaces of brick and stone made it easy for Bundler to The second test subject is a piece of masonry in the grounds of
find distinctive points for matching, making standing masonry an Bury St Edmunds Abbey. This section of wall is extremely irregular
ideal candidate for SfM. Our case study looks at St Andrew’s Church, in shape and would be hard to capture accurately using plan
Jevington, in East Sussex, for which data, in the form of georefer- drawings (see Fig. 3). It is possible that this piece of masonry is a
enced points from total station survey and some elevation draw- part of the Saxon ‘St Benedicts Basilica’ as described by
ings, had previously been produced. Jevington was visited on June Whittingham:
25th 2011 and 244 photographs of the church tower were pro-
On one of the two lumps representing the Infirmary Chapel of St.
cessed successfully to create a pointcloud using Bundler and
Michael, dedicated by Archbishop W. Corbeuil (1123-36), a Norman
PMVS2. The georeferenced total station points used distinctive
external string-course below former N. windows abuts an earlier
points on the masonry of the church tower, and these had to be
wall. From the latter instead of a Norman cross-vault springs a
estimated and matched in the pointcloud. For each point used in
barrel vault running, astonishingly, N. and S. across an aisled nave
the transformation the distance between the surveyed point and
(Whittingham, 1951).
the transformed pointcloud point was calculated. The greatest
distance between the two sets of points was 4.4 cm. For an addi-
tional set of 5 control points the greatest distance was 8.3 cm. Given The barrel vault arches towards the west and undercuts much of
the difficulty of matching the features across both files, and the the masonry. This piece of Masonry had not yet been surveyed or
uncertainty of the exact location for which the measurements were planned, so before photographing it a series of points were marked
made, the largest discrepancies may well be caused by human out around the base of the wall using coloured circles, and their
error. locations were recorded with a total station, along with a number of
In order to be useful as a record of the building fabric, we must points on the wall itself to use later to check the accuracy of the
be able to see the features clearly. This was achieved by creating a pointcloud transformation. The circle markers were photographed
polygon mesh from the pointcloud. The colours of the points were along with the masonry so they would be visible in the pointcloud
used to colour the mesh, and the result is a realistic representation (Fig. 3a). 453 photographs were taken of this piece of wall. The
of the exterior walls in 3d. The mesh was created using a Poisson camera was mounted on a pole to take shots of the wall from above.
Surface Reconstruction filter (Meshlab, 2013) (Fig. 1). Having the markers visible in the pointcloud made georefer-
In Fig. 2 an orthographic representation of the south wall of the encing simple. All the markers that were used for the trans-
tower can be compared to the elevation drawing. The red points formation were placed on the ground rather than the wall itself so
represent the georeferenced points from total station survey. The as not to obscure the masonry. To check that the pointcloud was

Table 1
The key steps in the SfM process for different toolchains.

Software Low density pointcloud High density pointcloud Georeferenced mesh and texture

Bundler & PMVS2 Using one of a number of available A second script uses the bundler output to Further processing must be done in other software
scripts (eg. ArcheOS) to point at a remove distortions from the photographs such as CloudCompare (to georeference points) and
set of photographs, the bundler and launch the PMVS2 process. The result Meshlab (to create and texture meshes)
pipeline returns a text file containing is one or more .ply files containing a high
camera locations, parameters and a density pointcloud
low density collection of points.
VisualSFM & PMVS2 VisualSFM has a GUI that allows a set PMVS2 can be launched from within the Further processing must be done in other software
of photographs to be loaded, and takes VisualSFM GUI. The result is a series of .ply such as CloudCompare (to georeference points) and
the user through a series of steps to pointcloud files very similar to those Meshlab (to create and texture meshes)
match points and find camera positions. produced by Bundler and PMVS2. These
The cameras and low density points are can be viewed in the GUI or opened in
displayed in a 3D viewer. an external editor such as Meshlab.
Photosynth A set of photographs is uploaded to the No further processing is currently possible
Photosynth website. The result is an
online 3D viewer which displays a low
density collection of points.
123D Catch A series of photographs are uploaded to the 123D
Catch website and the result is a fairly low resolution
textured 3D model that can be downloaded.
Georeferencing can be taken from points on the
images.
Photoscan A set of photographs is loaded into the The next stage in the process produces a Further processing within Photoscan will produce
software and when processed the camera high density pointcloud which can also be a model and a high resolution texture. It also has the
positions and low density points are displayed in the viewer. facility to georeference points.
displayed in a 3D viewer
176 S. Green et al. / Journal of Archaeological Science 46 (2014) 173e181

Fig. 1. Comparing the SfM model of Jevington Church to a photograph.

scaled correctly in all dimensions some distinctive points on the mesh we can also use the 3d model to examine cross sections.
wall were also surveyed. It was estimated that it should be possible Fig. 4 shows the mesh sliced parallel to the ground plane (4a) and
to pinpoint these to within approximately the nearest 3 cm. When perpendicular to the north/south axis (4b to f) This allows us to
the difference between the georeferenced pointcloud and the see the curve of the barrel vaulted roof described by
control points was calculated the greatest distance between them Whittingham (1951). The plans also suggest the presence of a
was found to be 2.9 cm. blocked window or an altar-niche at A, shown in cross section in
After creating a mesh (using the same filter as above), we can 4e, and a possible curve of a doorway at B (Shapland, 2012). The
plot elevations for the front, back and sides, and in this case also cross sections also reveal useful and hitherto under-appreciated
the top of the wall. However to fully understand the shape of the information for conservation of this structure: in particular, the

Fig. 2. A comparison of the 3d model and planned elevation overlaid with georeferenced points.
S. Green et al. / Journal of Archaeological Science 46 (2014) 173e181 177

Fig. 3. a. Photograph of the wall showing survey markers, b. Bundler cameras and pointcloud, c. PMVS2 dense pointcloud including survey markers.

masonry is less solid than it appears at first, being extremely thin pointcloud as a few pixels each of contrasting colours. The greatest
in some sections. discrepancy between the location of the markers and the trans-
formed points is 0.318 m.
4. Cases studies: digital elevation models A DEM was created from the pointcloud. This is relatively easy
as, for example, a commonly used pointcloud format such as .ply
4.1. Windmill Hill contains a simple list of coordinates followed by colour values,
separated by spaces which can be read into an open source GIS and
In order to investigate the accuracy of GIS data created using interpolated to create a DEM (e.g. GRASS GIS r.in.xyz followed by
SfM we need a high resolution digital elevation model for com- r.resamp.rst). In addition to the DEM, we can create a matching
parison. The solution is to find an archaeologically rich landscape raster dataset which summarises the colour information from the
for which there is LiDAR data available. Windmill Hill near Avebury photographs. As the markers were surveyed to a local grid, a further
in Wiltshire has existing LiDAR coverage at 0.5 m resolution, the transformation was needed to match the SfM DEM to the LiDAR
highest commercially available, and it has distinct archaeological data. This was done with an affine geographic transformation (first
earthworks. There is a Neolithic causewayed enclosure with three order polynomial) using two points, the elevation data was also
concentric ditches. There are four bronze age barrows within the matched at the highest point on both maps (but not scaled).
area of the enclosure and several more outside (Keiller and Smith, Fig. 5a and b show the area of Windmill Hill that is covered by
1965). Windmill Hill was photographed in the first week of July both maps. It is immediately apparent that there are problems with
using a camera rig attached to a kite. High winds meant that the the SfM map. The elevation values are patchy, and there appears to
camera was constantly moving, so a very high shutter speed had to be a fault line running unevenly from the north east to the south
be used to prevent motion blur. Dark conditions meant using a large west. However before looking at the problems evident in this map
aperture and high ISO resulting in slightly grainy images. 494 it is worth considering where it has succeeded. The geographical
photographs of acceptable quality were taken over a period of coordinates are so similar that the two maps can be overlaid
about an hour and a half, before it became too wet to continue. The without any blurring of the features. The height data is also largely
area covered was roughly a third of the causewayed enclosure, correct, with the average values across the maps being very similar.
about 250 m by 250 m. So what has caused the fault line? Most likely, we are looking at a
As well as photographing Windmill hill, a series of coloured combination of two factors, the impact of wind on image capture
markers of 15 cm diameter were surveyed using a total station. The and the presence of vegetation. The grass on the hill was several
markers are visible in the photographs and show up in the inches long when it was photographed and there were large

Fig. 4. Cross sections of the wall. The first image shows where the sections are made.
178 S. Green et al. / Journal of Archaeological Science 46 (2014) 173e181

vegetation. The feature at B is equally clear in both elevation models


because a path runs along this line, so the vegetation has been
trampled down to ground level. Unfortunately there is no way to
remove the influence of vegetation from SfM data unless the
vegetation is very thin and the ground is clearly visible beneath.
To learn more about the accuracy of the height data, it is worth
looking at a smaller area around the largest barrow, near the crest
of the hill. A colour map was derived using the RGB values of the
points (Fig. 6a) and the height of the DEM was matched to the
LiDAR DTM at the highest point of the barrow as this was free of
vegetation. If we subtract the SfM DEM from the LiDAR data we are
left with a map showing the difference between the two (Fig. 6b).
The white values are the areas of the DEM that are between
0 and 30 cm higher than the LiDAR data. Most of the map is within
these values, which would be expected if the difference is due to
vegetation. Very few areas have a lower value (pink). The dark pink
patches are most likely to be due to a slight geographical offset
between the two sets of data. If we compare the areas where the
DEM is over 30 cm higher to the SfM RGB map in Fig. 6 we can see
that these correspond to the darker patches of nettles. It is also
apparent from the colour map, which has not been reduced to a
lower (0.5 m) resolution to match the LiDAR data, that the SfM
maps offer potentially much higher resolution (i.e. in terms of ob-
servations per unit area).

4.2. Fishbourne Dolphin Mosaic

Fishbourne Roman Palace is a large Roman Villa near Chichester


in West Sussex. Excavations in 1960 unearthed a large number of
mosaics dating from the first to third centuries AD. The mosaics
Fig. 5. a. Structure from motion DEM at 0.5 m resolution, b. LiDAR DTM at 0.5 m now form part of a museum which was constructed in situ,
resolution Ó Environment Agency copyright 2013. All rights reserved.
covering over the whole of the north wing of the villa (Cunliffe,
1971). The ground underneath the ‘Cupid on a Dolphin’ mosaic is
patches of nettles covering parts of the hill. The vegetation itself can subsiding, and the mosaic is gradually collapsing with it. Fish-
account for the smaller variations in height, but not for the fault bourne museum were interested to see if structure from motion
line. It is likely that the latter was caused by the movement of the could be used to create a detailed DEM of an individual mosaic
vegetation by the strong winds, which allowed Bundler to match which could be used to monitor further subsidence in the future.
features that had moved between images. The LiDAR data to which The mosaic is too fragile to stand on, so in order to get detailed
we are comparing the DEM is a Digital Terrain Map, or DTM which photographs of the centre the remote controlled camera rig for the
has been post-processed to remove the vegetation (Crutchley, kite was hung from a rope stretched across the mosaic. The results
2010). Therefore, comparing this with the SfM DEM is not strictly were obtained from 653 photographs. The amount of detail in the
comparing like with like, and some discrepancy between the two is pointcloud is best illustrated by the untextured 3d model created
to be expected. The influence of the vegetation can be demon- from the points (Fig. 7). It is possible to make out the broken edges
strated if we compare the DTM and DEM in Fig. 5a and b. The of the tiled areas, and in some places individual tiles.
features marked with an A are the ditches of the causewayed camp. To georeference the pointcloud and assess the accuracy of the
They are clearly visible in the LiDAR DTM, but almost completely DEM, 46 points on the mosaic were surveyed using a total station.
invisible in the SfM DEM because they are overgrown by The use of a mosaic as a subject presents an opportunity to

Fig. 6. a. Colour map at 0.2 m resolution, b. The difference between the SfM DEM and LiDAR DTM.
S. Green et al. / Journal of Archaeological Science 46 (2014) 173e181 179

world points and the transformed pointcloud points and presents


these results as a percentage of the pointcloud size for each study.
This table includes data from the 3 cases studies described above, as
well as three further studies of very similar type.
The mean difference for the case studies is between 0.549% of
the pointcloud (Edinshall) and 0.039% of the pointcloud (Fish-
bourne.) The case studies have already suggested that there is a
correlation between the discrepancy in results and the degree of
error introduced when measuring or matching the points. To
investigate whether the correlation is real an estimate has been
made of the degree of error likely in either the measurement of the
real world coordinates or in the selection of matching points in the
pointcloud. These are documented in Table 3.
To assess the correlation between these two sets of points the
mean distances between points were plotted against the estimated
Fig. 7. Fishbourne Dolphin Mosaic untextured model. errors, and the linear regression was calculated using the statistical
program R. The adjusted R2 value for this relationship is 0.9591
which suggests a very strong correlation with a very significant p
georeference points without using markers. Instead tiles in value of 0.0000737 (calculated using student’s t.)
distinctive positions were chosen and noted on a plan of the Fishbourne mosaic has the smallest estimated error of mea-
mosaic. The individual tiles are approximately 1 cm across. As there surement (0.039%) and a large number of surveyed points to
were as few as 3 pointcloud points representing each tile the ac- consider, so further investigation was carried out on this dataset.
curacy cannot be much greater than the nearest 1 cm. The greatest Table 4 shows the elevation differences between the surveyed
difference between the surveyed and transformed points is 0.006 m points and same points in the DEM created from the SfM
which is well within our estimated error of 1 cm. pointcloud.
The largest difference is 0.006 m. This correlates well with the
5. Comparing the accuracy of SfM estimated 0.5 cm measurement error. However the value
of 0.006 m is quite common, so it would seem worth investigating
If SfM is to become a useful tool we must have some confidence further. Fig. 8 shows these values mapped onto the DEM.
in the accuracy of the results. It is impossible to achieve 100% ac- It is evident that there is a spatial pattern in these values.
curacy of measurement by any means so the calculations below Those in the central part of the mosaic are comparatively lower
attempt to assess whether the accuracy of the SfM pointcloud is than those at the edges. There are several possible reasons for
comparable to the alternative methods used in each case study. this. The most worrying would be that the pointcloud is not
Table 2 documents the distances between the georeferenced real- uniformly scaled, but this is highly unlikely because the z value is

Table 2
Highest and mean distances between observed (surveyed) points and their associated points in the georeferenced pointclouds.

No of points Greatest distance between Mean distance between Largest dimension Greatest distance Mean distance
tested observed and transformed observed and transformed of pointcloud as a percentage of as a percentage
points points pointcloud size of pointcloud size

Jevington 13 0.083 m 0.039 m 15.556 m 0.534% 0.251%


Bury St Edmunds 13 0.029 m 0.019 m 5.937 m 0.488% 0.320%
Bury Abbey 5 0.082 m 0.055 m 128.841 m 0.064% 0.043%
Edinshall 6 2.048 m 0.962 m 175.285 m 1.168% 0.549%
Windmill Hill 9 0.318 m 0.188 m 355.246 m 0.090% 0.053%
Fishbourne 42 0.011 m 0.005 m 12.734 m 0.086% 0.039%
Convoys Wharf 12 0.058 m 0.036 m 35.581 m 0.163% 0.101%

Table 3
Estimated errors due to measurement of coordinates and selection of points.

Estimated error Reason for estimate Estimated error as a percentage


in measurement of pointcloud size

Jevington 0.05 m Estimated variation between distinctive stones 0.321%


and the points representing them
Bury St Edmunds 0.025 m Distance between pointcloud points on the most 0.421%
poorly represented markers
Bury Abbey 0.075 m Radius of the marker, sometimes represented by 0.058%
a single point.
Edinshall 1.00 m Approximate average height difference between 0.570%
10 m cells of the OS DEM against which it is
georeferenced.
Windmill Hill 0.075 m Radius of the marker, sometimes represented 0.021%
by a single point.
Fishbourne 0.005 m Radius of a single mosaic tile, used as marker reference 0.039%
Convoys Wharf 0.03 m Radius of the marker, sometimes represented 0.084%
by a few scattered points.
180 S. Green et al. / Journal of Archaeological Science 46 (2014) 173e181

Table 4
Elevation values for the surveyed tiles subtracted from the DEM elevation values at the same points.

ID Diff ID Diff ID Diff ID Diff ID Diff ID Diff ID Diff

0 0.001 6 0.003 12 0.006 18 0.005 24 0.006 30 0.006 36 0.001


1 0.004 7 0.002 13 0.006 19 0.005 25 0.002 31 0.005 37 0.002
2 0.001 8 0.000 14 0.006 20 0.002 26 0.005 32 0.004 38 0.004
3 0.000 9 0.006 15 0.005 21 0.002 27 0.005 33 0.004 39 0.003
4 0.001 10 0.001 16 0.006 22 0.001 28 0.006 34 0.005 40 0.002
5 0.002 11 0.005 17 0.005 23 0.005 29 0.004 35 0.004 41 0.002

Fig. 8. The location of the surveyed tiles, and their elevation distance from the DEM.

meaningless when we consider the untransformed pointcloud, points (the precision) is accurate to a considerably higher reso-
and the affine transformation would not have changed this. The lution than 1 cm.
low values do not exactly match the lowest part of the mosaic, It seems that the Fishbourne pointcloud has higher precision
they are actually a better match for aspect; the lowest values tend than accuracy. In other words any errors are systematic across the
to be on the west facing slope. This discrepancy is therefore most whole pointcloud, and the point positions are correct to a higher
likely to be caused by a slight misalignment of the rotation or resolution when only their relative positions are considered. This
translation of the pointcloud due to the imprecision of the point means that if we can improve the accuracy of the measurements for
locations. If the transformation is the primary source of error it is the transformation we should be able to improve the accuracy of
likely that the relative position of the tiles is correct. This sug- the pointcloud. However to improve accuracy significantly we need
gestion is supported by the colour map. If we look at an enlarged more points in the pointcloud to match to the markers. This can be
section in Fig. 9, it is apparent that the relative position of the solved by increasing the resolution of the pointcloud or by using

Fig. 9. RGB georeferenced colour map. a. 1 cm resolution, b. 0.1 cm resolution.


S. Green et al. / Journal of Archaeological Science 46 (2014) 173e181 181

higher resolution photographs, but will lead to a steep increase in photographing of the dolphin mosaic at Fishbourne Roman Villa.
processing time. A better solution would be to pick out the centres Andrew Dunwell kindly made his survey data available. Giulia De
of the markers from the original photographs and ensure that a Nobili, Sophie Butler and James Mason provided technical assis-
point is created in the correct place. If this information was avail- tance during the collection of data.
able from the start of the Bundler process the pointcloud could be
created with the correct coordinates from the start.
References
6. Discussion Agisoft, 2013. Agisoft Photoscan Professional Edition. http://www.agisoft.ru/
products/photoscan/professional/ (accessed 11.01.13.).
This paper has demonstrated that structure from motion can be a ArcheOS, 2011. What Is ArchaeOs?. http://www.archeos.eu/wiki/doku.php?id¼start
(accessed 18.09.11.).
very useful tool for archaeology, and is likely to become more Autodesk, 2014. 123D Catch. http://www.123dapp.com/catch (accessed 09.01.14.).
important in future as it develops and addresses its failings. As it Crutchley, S., 2010. The Light Fantastic: Using Airborne Lidar in Archaeological
stands at the moment it is most useful as a less accurate, but cheaper Survey. English Heritage, Swindon.
Cunliffe, B.W., 1971. Excavations at Fishbourne, 1961e1969. Society of Antiquaries,
and higher resolution substitute for some of the cutting edge tech- London.
nology used in archaeology, such as LiDAR and laser scanning, or as a De Reu, J., Plets, G., Verhoeven, G., De Smedt, P., Bats, M., Cherretté, B., De
supplement to photogrammetry, topographic survey and plan Maeyer, W., Deconynck, J., Herremans, D., Laloo, P., Van Meirvenne, M., De
Clercq, W., 2013. Towards a three-dimensional cost-effective registration of the
drawing. This places it in an important strategic position, occupying archaeological heritage. J. Archaeol. Sci. 40, 1108e1121.
the middle ground between technology that is prohibitively Ducke, B., Score, D., Reeves, J., 2011. Multiview 3D reconstruction of the archaeo-
expensive and traditional techniques that are slower and of lower logical site at Weymouth from image series. Comput. Graph. 35, 375e382.
Furukawa, Y., Curless, B., Seitz, S.M., Szeliski, R., 2010. Towards Internet-scale Multi-
resolution. Until structure from motion can demonstrate reliable
view Stereo. Computer Vision and Pattern Recognition (CVPR), pp. 1434e1441.
accuracy, and this can be calculated on a case by case basis, it is Girardeau-Montaut, D., 2014. CloudCompare. http://www.danielgm.net/index.php
unlikely to be taken as seriously as a measurement tool, however it is (accessed 09.01.14.).
a technology still in its infancy, and this problem is understood by its Keiller, A.F.S.A., Smith, I.F., 1965. Windmill Hill and Avebury. Excavations by Alex-
ander Keiller, 1925e1939 (Prepared for Publication by I. F. Smith.) [With Plates,
developers. Moreover, SfM brings with it other benefits, in particular Including a Portrait, and Maps.]. Clarendon Press, Oxford.
the high speed and simplicity of acquisition, which allows it to be Lowe, D.G., 2004. Distinctive image features from scale-invariant keypoints. Int. J.
used in situations where slower or unwieldy technology would be Comput. Vis. 60, 91e110.
Meshlab, 2013. Meshlab. http://meshlab.sourceforge.net/ (accessed 11.01.13.).
inappropriate, allowing, for example, the capture of models of fea- Microsoft, 2014. Photosynth, http://photosynth.net/ (accessed 09.01.14.).
tures being excavated without slowing down the excavation. It re- Olson, B.R., Placchetti, R.A., Quartermaine, J., Killebrew, A.E., 2013. The Tel Akko
sults in data that is more complete than any other technology, as it Total Archaeology Project (Akko, Israel): assessing the suitability of multi-scale
3D field recording in archaeology. J. Field Archaeol. 38, 244e262.
can capture both shape and colour of uneven surfaces at a high Shapland, M.G., 2012. Buildings of Secular and Religious Lordship: Anglo-Saxon
resolution. Structure from Motion promises to become a highly Tower-nave Churches (unpublished PhD thesis). University College London.
important tool in its own right if it is developed correctly. For this Snavely, N., Seitz, S.M., Szeliski, R., 2006. Photo Tourism: Exploring Photo Collec-
tions in 3D. ACM, pp. 835e846.
reason it is very encouraging that so many of the developments in Szeliski, R., 2010. Computer Vision: Algorithms and Applications. Springer.
SfM have remained open source, which gives archaeologists a chance Verhoeven, G., 2011. Taking computer vision aloft e archaeological three-
to mould this technology to suit their own archaeological needs. dimensional reconstructions from aerial photographs with photoscan.
Archaeol. Prospect. 18, 67e73.
Verhoeven, G., Doneus, M., Briese, C., Vermeulen, F., 2012. Mapping by matching: a
Acknowledgements computer vision-based approach to fast and accurate georeferencing of
archaeological aerial photographs. J. Archaeol. Sci. 39, 2060e2070.
This dissertation owes a great deal to the help of many people. Whittingham, A., 1951. Bury St. Edmunds Abbey: the plan, design and development
of the church and Monastic buildings. Archaeol. J. 108, 170e171.
Benjamin Ducke and Mark Lake provided their help and guidance Wu, C., Agarwal, S., Curless, B., Seitz, S.M., 2011. Multicore Bundle Adjustment.
with the writing of this paper. Robert Symmons arranged for the Computer Vision and Pattern Recognition (CVPR), pp. 3057e3064.

You might also like