1 5m 
  
  
Figure 5. Digital plan view and side views of textured model with over imposition of traditional survey. 
housing, cheap software for digital photogrammetry) it wasn't 
possible to use this academic way. It was not possible to modify 
the colinearity equation in PhotoModeler and here was no 
possibility to get a professional housing for our cheap camera 
(Nikon coolpix 990) The housing we used was from Ikelite TM 
and had a special lens for which the calculation assume water in 
the object space. The use of wide angle was then possible. In 
these condition, according the approximate pin-cushion 
distortion model from [Kwon, 1998] and keeping in mind that 
for a survey of piece of marble laiyngunder the water since 
almost two thousand years we weren't dealing with millimetres 
in term of precision, we decide to use a very simple way to 
calibrate the camera : the PhotoModeler calibration module. 
We just considered the set housing-camera rigid and 
undeformable and we made a calibration as if we were in air. 
The pin-cushion distortion is here modelized by focal length 
variation (increase) and radial distortion. 
3. PLOTTING PHASE 
3.1 Photogrammetric restitution 
The photogrammetry application carried out on the marble and 
dolia wrecks, has been realized on using a known methodology 
for underwater survey, taking into consideration some specific 
characteristics of this experiment: in the first case we used the 
Photomodeler software for the restitution of the marble blocks 
with 2 degrees of approximation; in the second case we used the 
Arpenteur software [Drap, Grussenmeyer, 2000], suited to the 
specific restitution of the dolia. The complexity of elements to 
survey and their scarce recognition have imposed a procedure 
of photogrammetric restitution, subdivided in phases made out 
as follows: 
1) The first phase, necessary to establish a relative orientation, 
has been realized on defining a net of homologous points, 
(between 12 and 18 points for each orientation), referencing at 
least 4 adjoing photographs (2 shooting in succession and other 
2 on the near mark); this grid of 400 points does not correspond 
directly to the final model, given that it does not define in detail 
the geometry of the surveyed marble elements. 
2) In the second phase the absolute orientation of the model has 
been determined using rulers set out on the field and the 
plumb-line of some floats as a dimensional landmark, anchored 
on the bottom, as vertical axis *Z", on which the model can 
whirl. 
3) In the third phase the geometrical conformations of each 
single element have been marked after orienting all the 
photographs on indications given by the traditional survey, 
realized by the Soprintendenza of Tuscany.In other words once 
established the formal-dimensional characteristics of an 
element, for instance a beam with rectilinear surface or a 
column with a curvilinear trend, it has been necessary to mark 
aligned points on transversal and representative sections. In the 
first case, these points identify a section with rectangular trend 
(minus lacks due to erosions and surface concretions) and in the 
second case, they prove a section with circular trend. 
3.2 3D plotting 
Subsequently, the model has been re-defined in all its parts of 
surface discontinuity (in origin corners, breaks or contacts 
between different elements), constituting a 3D grid of about 
3000 points, useful for texturing the model, with a final 
approximation of 0.027 percentage. Such operation was 
realized, considering the difference between the object in its 
original conformation, estimated by “anastilosi” and the 
restored model. 
The greatest problems regarded principally the covering of the 
model in its entirety, and the homogeneity of points’ 
correlation. In the first case not having shot photographs in 
excess regarding the external perimeter, we didn’t manage to 
have any photograph which could make a stereoscopy of the 
outside faces; regarding the second case, instead, being the 
—330— 
  
fldmb a udi abe sw i