-15- 
The final photogrammetric model was then generated with the 
Stereoview software supplied by Nikon. The relative orientation 
gave CT=2pm, while the global accuracy, checked with the 
topographic target coordinates, was around 1mm. 
4.2.3 Off-the-shelf camera: Photogrammetry was also 
performed with an off-the-shelf camera coupled with a software 
for post processing the acquired images. 
The camera used is a Nikon coolpix 5000, equipped with a 2/3" 
CCD organized in 2560x1920 pixel (5.2 MPixel), and a zoom 
lens that was kept fixed, during the photogrammetry shots, at its 
shorter focal length (7.1 mm), equivalent to a 28 mm wide- 
angle lens in a 35mm reflex camera. The camera has also the 
capability of saving images in uncompressed tiff format. This is 
an important feature when the image is employed for 
measurement purpose, because in this way any processing over 
the image can be avoided. 
The photogrammetry program is a commercial package (Shape 
Capture - Shape Quest Inc., Ontario, Canada) that allows 
calculating camera calibration with simple steps. 
After calibration, a set of convergent digital pictures of the 
object taken from different angles can be registered by first 
selecting homologous points over them, and then launching a 
bundle adjustment procedure (El-Hakim et al., 1996). 
Once the images are registered, the 3D coordinates of specific 
points can be easily generated and saved in a text file. 
4.4 TOF based Scanner 
The smaller scale 3D acquisition was performed with a TOF 
scanner LMS-Z420 (produced by Riegel, Austria), giving a 
measurement uncertainty of ±8 mm. 
4.5 Triangulation based Scanner 
At larger scale a range camera “Opto3D Ranger” was employed 
(produced by Optonet, Italy). It is a triangulation system based 
on projection of light patterns over the investigated surface. The 
main components of the system are a LCD projector for pattern 
projection, a C-mount CCD camera, a few camera lenses, a 
variable baseline support, and a calibration fixture. Depending 
on the type of surface imaged, the field of view and accuracy of 
the 3D camera are reconfigured. This process allows the 
operator to adapt the modeling process to different situations 
found in modeling of artworks. 
The 3D coordinates evaluation in this system is based on the 
simultaneous use of Gray codes projection, for generating an 
approximate estimation of the surface, and of phase shift of the 
finest pattern for a refinement of each 3D point estimation, 
according to a method described in the literature. 
As any other active optical sensor based on triangulation, the 
measurement uncertainty depends directly on the projector- 
camera distance (baseline) and inversely on the camera-field-of- 
view-distance (standoff). 
Since the displacement between the 3D points measured by the 
scanner and their true coordinates follows a gaussian 
distribution, it is common to statistically quantify the 
measurement uncertainty through the standard deviation (ct x , a y 
and ct z ) of such displacements respect to the three reference 
axes. 
This evaluation is performed at the calibration stage, usually 
before starting a measurement session, by measuring the 
position in space of a known target located in known positions. 
It is also made in the following stages, for example at the end of 
a measurement session, or at the beginning of a session where a 
previous calibration setup is employed, by acquiring a 
verification object with a highly planar surface, and evaluating 
the deviation of the point cloud from the corresponding best 
fitting plane. 
In the Results section, the opto-geometrical setup chosen for the 
scanner gave a field of view of 290x210 mm on the focal plane, 
with an extent of 160 mm in depth, and a standard deviation 
along z (cj z ) equal to 73pm. 
5. EXPERIMENTAL RESULTS 
The topographic survey has included some known points, 
measured in previous surveys, performed since 1909 by the 
Istituto Geografico Militare (IGM), and afterwards (1972-73; 
1994) by the group of Prof. Fondelli at the University of 
Florence for measurements over the cathedral dome. Such 
coordinate system has been taken as an absolute reference for 
the work described in this paper, and all the following 
measurements have been re-oriented according to it. 
Although the topographic survey gives the better possible 
accuracy over some specific points, for the mean part of the 
examined structure we chose to use photogrammetry due to 
logistic reasons (the place located underground is accessible 
only by a small aperture in the baptistery floor and the light is 
very little), and for the need of filling the gaps between accurate 
points well apart each other. 
Finally photogrammetry was used to quickly generate a 
reference system for 3D acquisitions. 
The procedure for integrating range maps originated by 3D 
scanning and digital photogrammetry is based on specifically 
designed targeting plates, capable to be properly identified with 
both measurement techniques, to be placed in the scene. 
The main role of these plates was to be easily recognizable with 
both 3D scanning and photogrammetry, so they had to exhibit 
specific 3D and optical properties. For this reason we designed 
them in order to have: 
• high contrast for making easier the target selection with the 
photogrammetry software; 
• high planarity, so that during the post processing of range 
maps its surface could be approximated with a fitting plane 
instead of employing the raw data set, typically affected by 
errors in the high contrast areas of the image; 
• high number of targets for making more robust the 
rototranslation procedure. The minimum number of points 
to univocally calculate a rototranslation of the range-map 
are of course three, but in this way the possible errors 
affecting the rototranslation matrix would be directly 
influenced by the uncertainty of each point measured in 
both the source and destination reference system. By 
properly choosing and increasing the number of points to 
be employed for calculating the rototranslation, the 
uncorrelated errors tend to cancel each other, leaving only 
systematic errors, that can be minimized thanks to 
calibration; 
• a known distance between targets, in order to give us the 
possibility to check the accuracy and deviation of the target 
position measured with both systems, in any stage of the 
processing. 
The measurements set-up is shown in figure la, were the targets 
located over the mosaic are shown, in positions appropriate to 
get tridimensionality from the digital pictures of the scene, and 
in order to cover as much as possible of the framed area. In 
figure lb the groups of coordinates generated through the 
bundle adjustment of 4 digital images similar to figure la, and 
including the targeting plates A, B ,C and D in a single 
coordinate system, are shown.