-66 
2. GENERAL OVERVIEW OF THE PROCESSING 
PIPELINE 
3. TECHNOLOGY USED FOR THE BYZANTINE 
CRYPT 
The typical processing pipeline used for 3D modeling includes 
calibration/verification, geometric modeling, and appearance 
modeling. The sequence of steps required is well documented 
in (Soucy et al., 1996a). Here we summarize some of the steps 
for the reconstruction of a complete fully textured model. The 
calibration of a range camera is concerned with the extraction 
of the internal parameters of the 3D camera. The manufacturer 
should include with their commercial 3D camera a test object to 
verify the accuracy (verification). For 2D cameras, different 
calibration methods exist spanning from the simple pinhole 
model to the complete photogrammetric solution. 
Geometric modeling is essential to recreate realistic models of 
sites that have a lot of surface details difficult to model with 
photogrammetric techniques. In most cases, the creation of a 
3D model will rely on multiple scans (range images), taken at 
various locations all around an object or inside a site, that need 
to be registered. A few techniques have been devised for this 
problem. One method combines photogrammetry and laser 
range-imaging techniques (El-Hakim et al., 1998). Another 
method uses only the surface data from the multiple views 
(Soucy and Laurendeau, 1995). The views must have enough 
overlap between them to find the registration and to merge 
them together. One important feature of this approach is that 
the geometric details of the surface of the object itself are used 
to register the views together. Obviously, quasi-planar or 
spherical surfaces should be avoided with this latter technique 
and for those cases, the former method, or the following one, is 
recommended. Some 3D camera manufacturers have adopted 
the use of geometrical objects like spheres placed near a surface 
to be acquired in order to facilitate the registration between the 
individual 3D images. 
Appearance modeling includes methods like image perspective 
techniques (IPT) and reflectance modeling. IPT is concerned 
with direct mapping of photographs onto a 3D model 
(Weinhaus and Devarjan, 1997; El-Hakim et al., 1998; 
Neugebauer and Klein, 1999; Sequiera et al., 1999; Stamos and 
Allen, 2000). Reflectance modeling is used to extract from the 
measured colour and shape those physical properties of an 
object that are intrinsic to it and that determine its appearance 
when viewed with artificial lighting on a computer screen 
(Baribeau et al., 1992; Bemardini et al., 2001). Texture 
mapping is also an efficient way to achieve realism with only a 
low resolution, faster to render, geometric model. Recently, 
techniques that map real-scene images onto the geometric 
model, also known as image perspective techniques (IPT) have 
gained a lot of interest. Though, some commercial 3D systems 
supply a colour texture in registration with the 3D image 
unfortunately with very limited visual image quality. Hence 
separate cameras acquire high-resolution colour images, which 
can be precisely mapped onto the geometric model provided 
that the camera position and orientation are known in the 
coordinate system of the geometric model. The main challenges 
faced by people in that field are adequate lighting, accurately 
computing lens distortions, 2D camera to 3D-model pose 
estimate, dealing with hidden surfaces and incomplete views 
(El-Hakim et al., 1998). 
To model the Byzantine Crypt (Fig.l), we chose a 
photogrammetric technique for the outside (i.e. main and 
secondary entrances located above the Crypt) and a laser range 
scanner that provided plain clouds of 3D points for the Crypt 
(located underground). Texture information was not available 
from the scanner and therefore, it was acquired separately with 
a high-resolution digital camera. Two-dimensional imaging is 
not only used to record appearance but also to perform 
geometric measurements and to produce 3D textured models. 
Proper camera calibration and bundle adjustment algorithms 
combine in digital photogrammetry to give accurate feature 
coordinates and reliable pose estimations (Triggs et al., 2000). 
Many commercial packages perform this task quite nicely 
(Debevec et al., 1996; El-Hakim, 2001). The model of the main 
entrance is shown on Figure 2. The model for the secondary 
entrance was built in a similar way (Figure 3). 
a) b) 
Figure 2. Crypt main entrance build with photogrammetry, a) 
textured model, b) wire frame model. 
a) b) 
Figure 3. Crypt second entrance build with photogrammetry, a) 
textured model, b) wire frame of model. 
We selected a SLR-type digital camera, the Nikon Dlx for the 
texture acquisition. The CCD sensor has an area of 23.7 x 15.6 
mm and an effective pixel count of 4028 x 1324. The output 
image is re-interpolated to a resolution of 3008 x 1960 pixels 
(imager ratio 3:2). Both native (NEF) and TIFF formats are 
available. Proper texturing of the 3D model requires special 
lighting fixtures in order to control illumination. Good 
uniformity of the illumination is important in order to ease the 
processing tasks. In an environment with frescoes, the main 
problem with lighting is the amount of heat generated by high 
power lamps. In this case, the amount of heat must be kept to a 
minimum to avoid damage. Xe flashtubes with a colour 
temperature of about 5600 K were used. The tubes are UV 
coated and the stored energy is about 500Ws with duration of 
1/700 sec. The manufacturer rates the stability at ±1%. All of