at one time or another, mostly with field surveys commissioned
by the Ministry of Culture which has thus amassed a wealth of
planar data with no, or sparse, elevation data. Graphical data of
this type can be well exploited at an extended scale as exclusive
control information to generate orthomosaics for innumerable
sites, at least as a basic archival documentation. One may pro-
ceed by simply deriving planar control points from such 2D in-
formation and perform a purely planimetric adjustment. The re-
sulting inaccuracy in absolute orientation, however, could be di-
minished by exploiting the 3D model elevations with a suitable
weighting. The results presented here indicate that, in fact, this
pre-existing information may prove useful in several cases for
cost-effective approaches of orthophotography by limiting, or
even eliminating, the need for new field surveys. The aspects of
archaeological orthoimaging referred to above are discussed and
illustrated with examples from four different Greek sites.
2. PHOTOTRIANGULATION
This is a key issue in archaeological surveys since, for a number
of reasons already referred to, the questions of performing the
bundle adjustment and recovering reliable values for the image
exterior orientation parameters may well not be trivial. Starting
with interior orientation, it must be pointed out that a full self-
calibrating bundle adjustment may often be infeasible. This may
be due to a combination of the limited extension in depth of the
object, the inaccuracy and low ‘identifiability’ of control points
(which, more than often, need to be simple natural detail points)
and the unfavourable strip geometry. Full pre-calibration is one
way to tackle this problem. However, this is not always practic-
able. For instance, in each of the four projects which serve here
as examples a different non-metric camera has been employed
(all belong to the Department of Surveying & Photogrammetry
of the Greek Ministry of Culture).
In case of relatively limited relief, Karras and Mavromati (2001)
have demonstrated that the use of ‘nominal’ interior orientation
values (the principal point in an analogue camera is ignored; the
nominal focal length is used as camera constant) does not affect
accuracy to a considerable extent. Contrary to this, the effects of
radial lens distortion may be decisive, particularly in the case of
wide-angle photography. The correction of this error is capable
of trebling accuracy. Lens distortion could be modeled through
bundle adjustment, but may also be estimated separately by em-
ploying simple techniques of partial pre-calibration, for instance
using images of straight linear features. This approach has been
adopted in all examples used here, which form parts of wider
projects. These are the following:
e Sparta (parodoi walls of the ancient theatre in Sparta). The
strip used here consisted of 6 images of mean scale 1:250, taken
with a medium format Mamiya camera with wide-angle 45 mm
lens. Recording has been performed with horizontal camera axis
using a fishing-rod to raise the camera. For the 44 control points
used, the RMSxyz error of bundle adjustment was 1.6 cm.
* Aigosthena (eastern facade of the ancient castle in Aigosthe-
na). Here again a total of 6 images acquired with horizontal axis
were used. The medium format Fuji camera (45 mm wide-angle
lens) was raised with a small meteorological balloon to give a
mean image scale of 1:300. The RMSxvz error for the 68 control
points used was 1.3 cm.
* Zea (small ancient theatre in Piraeus). The 4 images used in
this instance had been acquired vertically, employing the same
means, with a small format Cannon camera (28 mm wide-angle
lens). For a mean image scale of 1:1200, the RMSxyz error for
82 control points was 2.8 cm.
e Ag. Marina (archaic site in Athens, dedicated to Zeus). The
7 images selected here had a mean scale of 1:1100 and had been
acquired vertically as above with a small format Nikon camera
(28 mm wide-angle lens). The RMSxyz error for the 100 control
points was 3.7 cm. [This project has been fully documented in
Karras et al., 1999.]
To a considerable extent, the satisfactory RMS errors referred to
above are attributed to the correction of radial distortion. How-
ever, further aspects of a successful adjustment have also to be
mentioned. With the means for raising the camera used, ‘flight’
planning cannot be fully adhered to, which may result in rather
unfavourable imaging geometry. In vertical photography, image
tilt does exist but apparently can be limited below 5?. The pro-
blem here are mainly the differences AK in rotations about the
vertical camera axis, which may even exceed 15°. Contrary to
this, horizontal photography suffers mainly from ¢-tilts about
the vertical image axis (which in the case of Aigosthena exceed-
ed 15?). To confront this problem, relatively small stereo-bases
are required to secure adequate overlap, along with liberal con-
trol information of sufficient accuracy and tie points determined
by as many rays as possible. Control and tie points, which often
are not signalised (as in the examples discussed here), must be
measured carefully on the image, particularly if significant per-
spective distortion is present as a consequence of surface relief
and image tilt. Actually, it is this need for ample ground control
which has led to investigations, presented in Section 4, regard-
ing the possibility to exploit pre-existing plans as a source for
ground control. Finally, the differences in imaging distances are
also to be kept within certain tolerances (if image resizing and
processing, for instance, with different camera constants or strip
segmentation in smaller parts is to be avoided).
3. SURFACE MODELING
As mentioned already, accurate surface modeling is a key issue
in the generation of orthoimages both geometrically reliable and
visually correct. Locally inaccurate description of very demand-
ing surfaces leads here to geometric inaccuracies and 'stretched'
or *melted' orthoimage areas. The commercially available soft-
ware commonly used represents object surfaces as a DTM with a
single value Z for each planimetric XY location (more complex
surfaces not representable in this way call for special treatment;
e.g. Knyaz & Zheltov, 2000). All photogrammetrically collected
elevation points and breaklines are typically integrated by De-
launay triangulation into a surface mesh defined by triangles. In
fact, manual stereoscopic measurement is still the main mode of
collection. Automatic DTM generation in archaeological ortho-
imaging remains an open question (Baratin et al., 2000). Laser
scanning collection, on the other hand, faces problems of post-
processing for the creation of triangulated meshes suitable for
the existing orthophoto software (Bóhler et al., 2001). Besides,
not every archaeological site is accessible to laser scanners as it
may be to photography.
Obviously, attention must be paid in the collection phase to the
inclusion of all significant surface breaks and discontinuities (a
process which also requires certain amounts of experience). But
one must also a priori have a clear idea of the type of algorithm
which will be used to generate the surface model for orthophoto
production (as pointed out by Baratin et al., 2000). In the expe-
rience of the authors, perhaps the most usual problem in ortho-
imaging archaeological objects is modeling surfaces orthogonal
to each other, i.e. the formation of ‘vertical’ triangles, a task en-
countered in all projects outlined above. In such cases, the soft-
ware needs to be ‘assisted’ by suitable collection.
Generally, data for vertical faces are sampled as a combination
of breaklines on top with spot heights at the bottom. This, how-
ever, does not necessarily protect from a ‘random’ triangulation
which will later cause a deformation during image resampling.
Attempting to create orthoimages possibly equivalent to the im-
portance of the monuments, the following collection scheme has
-414-
been
by a
colle
to its
This :
form:
and, |
tolera
not a
other
but n
result
In Fi
clearl
whicl
A sec
ject.
Figu
the sl
be se
in Fig
ing fi
the Z
