the images were acquired with a fixed focal length to ease 
calibration of intrinsic parameters and an f/22 aperture to 
produce a large depth of field. 
In order to create a dense 3D model of the Byzantine Crypt, a 
MENSI SOISIC-2000 scanner was used. Tame I summarizes 
the specifications of this laser range scanner. This laser scanner 
can acquire 3D images at a minimal distance of 0.8 m and at up 
to 10 m with a measurement uncertainty varying between 0.4 
mm and 2 mm (distance-dependent). Though the Byzantine 
Crypt is relatively large (16.5 m by 10 m by 2.5 m), we still 
wanted to model it with a fairly high spatial resolution. For this 
size environment, there aren’t a lot of range cameras on the 
market that could provide us with the desired level of spatial 
resolution and measurement uncertainty. In fact, these distances 
represent the transition between optical triangulation and time 
of flight technologies. 
b) 
Figure 4. Complete 3D model of the Byzantine Crypt shown 
with synthetic shading, a) view from the outside showing the 
two entrances of the complete 3D model of the Byzantine 
Crypt, b) a particular view of the stairs leading to the Crypt. 
Size 16.5 m x 10 m x 2.5 m, spatial resolution of 5 mm, range 
uncertainty of 1 mm and accuracy of 15 mm. 
In order to keep a quasi-constant spatial sampling on the 
surface of the walls, 3D vertical scans acquired at 2.5 m were 
used to build the 3D model. A sampling step of 5 mm was 
agreed upon in cooperation with an art historian. This gave an 
average scan time per 3D image of about 80 min. And for that 
standoff distance, the depth uncertainty was estimated at about 
0.8 mm (1 sigma). Figure 4 presents the complete 3D model 
that would appear if one could see through the ground. From 
this model, a floor plan was created and is shown on Figure 5. 
We tested two techniques to align the 3D images, the first 
based upon spheres positioned strategically in the scene and the 
second based on data driven alignment (based on ICP) followed 
by a global alignment. Results are not reported in this paper but 
demonstrated the advantages of the latter method. 
SPECIFICATION 
VALUE 
Field of View 
46° x 320° 
Standoff (mm) 
800 
Maximum range (mm) 
10 000 
Resolution (X) minimum mesh size 
0.1 mm per meter 
of range 
Z measurement 
0.3 @ 800 mm 
Uncertainty-1 ct (mm) 
0.4 @ 2500 mm 
Cooperative surface 
0.6 @ 4000 mm 
Data Rate (Hz) 
100 
Scanner size (cm 3 ) 
73 *21 *28 
Scanner weight (Kg) 
16.3 
Output data type 
Cloud of points 
without intensity 
information 
Table 1 SOISIC™ 2000 laser range scanner specifications 
Figure 5. Floor plan generated from an orthographic view of 
the 3D model of the Crypt showing its dimensions. 
4. PUTTING IT ALL TOGETHER: IPT MAPPING 
ONTO 3D 
Projects aimed at the construction of dense 3D and appearance 
models have become too numerous to be listed here. Each 
project tries to optimize some part or all of the modeling 
phases. What we had to deal with was a 3D model that did not 
have intensity (also know as reflectance channel) data attached 
to it. This model was created after the merging process 
(removal of redundancy in overlapped regions) and different 
resolution models were also created after compressing the 
polygons to appropriate spatial resolutions. The technique 
implemented in the commercially available software 
Polyworks 1 M is explained in (Soucy and Laurendeau, 1995). 
The methodology proposed is very flexible and within reach to 
non-experts. It uses commercially available software and a 
small program that combines the re-projection of the 3D points 
found in the un-textured model file (e.g. VRML format) onto 
the texture images to give the complete realistic-looking and 
geometrically correct 3D textured model. This last module will 
become part of a commercial package. The 2D camera does not 
have to be rigidly mounted on the 3D camera and therefore 2D 
images created from digital cameras can be mapped onto the 
3D model. These 2D images can be taken specifically for