CIPA 2003 XIX th International Symposium, 30 September - 04 October, 2003, Antalya, Turkey 
546 
Figure 3. Longitudinal apse section and one facade 
of the transept. There USM on vector map and on raster image 
rectification are shown 
high resolution scanning subjected to reduction procedure. 
Only for the transept, Cyrax 2500 laser (by Leica) has been used, 
with sampling step of about 1.5 cm; in this way all the transept has 
been covered. For more complex details, as for example, capital 
blocks, a sampling step of 6 mm has been set. 
In all, 20 range maps, with an overlapping of about 40% have 
been generated. This overlap has been performed not only to align 
3D images but also to cover undercuts and hidden zones. Laser 
scanner has been mounted on its tripod or simply placed on the 
scaffoldings, at various levels, to avoid great inclinations. 
3. FROM POINT CLOUD TO SURFACE 
Actually, 3D modeling of real free-form surfaces consists of the 
following steps: registration, pre-processing, mesh generation, 
post-processing, texturing. Range map registration has been 
carried out with Cyclone software (by Cyra). To obtain the 
transept complete model, 20 range maps have been registered, 
13 for right zone and 7 for left zone. In order to establish 
connection with the two parts, (interrupted by the apse) and 
in order to set a unique reference system for all type of data, 
specific targets, topographically measured, have been used as 
additional constrains in the range data alignment. These target 
are automatically recognized by the software which calculates the 
Figure 4. Laser scanner setup 
Figure 5. Example of range map registration: 
we can see target position, automatically determinated, used 
to rototranslate the model in the topographic reference system 
centre. The registrations have been performed in scan block; for 
each one, 5 or 6 targets were available, 3 for the rototranslation 
in the reference system and the other as control points. 
All following data processing has been done on portions of the 
entire model, to be able to manage acceptable file dimension and 
to reduce elaboration time. Also visualisation became modifiable 
in real time. At the end, all blocks have been reassembled in order 
to create a unique model. 
Figure 6. A view of the model, integrated, in apse area, 
with the photogrammetric restitution 
In the pre-processing phase we have removed unrelated elements 
to the object of survey. This is an operation only partly automatic, 
for instance, using on-distance based filter. To refine the selection 
of these elements, manual operation is required (Figure 7). 
Also, noise reduction enhances the scan quality by removing excessive 
speckling with statistic methods, as we show in figure 8. 
In figure 9, there is an example of clusterization of registered 
block of range map. The partition has been performed marking 
the boundary of significative architectonic portions of about one 
million of points. 
The conversion of the measured data into a consistent polygonal 
surface implies the generation of triangular surfaces (FigurelO), 
satisfying some quality requirements, and mesh optimization 
Figure 7. Examples of unrelated elements