The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
penetration causes systematic shift - both vary with marble age, 
erosion, and surface dirt, but actual values are difficult to 
determine due to lack of complete understanding of surface 
response. Thus it is hard to completely correct this problem. 
2.2 Data Processing 
The huge data is impossible to process at the acquired high 
resolution, yet processing at lower resolution affects accuracy 
of operations such as registration. Reducing data size with 
simplification technique must ensure no loss of important 
details. Also combination of data taken by different sensors at 
different resolution, accuracy, and viewpoints affect the overall 
model accuracy if the quality of the different data are not 
properly considered. Despite using several sensors, some gaps 
and holes remained. This raises an important question: should 
we fill those with interpolated, but possibly inaccurate, surface 
patches or leave them out even though they may be visually 
unpleasant? One solution is to fill the gaps but keep accessible 
record of those uncertain filled areas. 
2.3 Realistic Appearance 
Photo-realism, defined as having no difference between the 
view rendered from the model and a photograph taken from the 
same viewpoint, goes much further than simply draping static 
imagery over geometry. Due to variations in lighting, surface 
specularity and camera gain settings, sensed colour and 
intensity for a segment shown in images taken from separate 
positions will not match. This is particularly problematic on 
large open-air site like the Acropolis. Also, measurement of 
surface reflection properties must be included for proper model 
lighting. However, the texture images contain whatever 
illumination existed at imaging time. Ideally this illumination 
should be removed and replaced by dynamic illumination 
consistent with the rendering point of view. Another problem is 
that the range of brightness in the scene cannot be captured in a 
single exposure by current digital cameras. This causes loss of 
details in the dark areas (shadows) and saturation in the bright 
areas (sun) if both coexist in the scene. It is thus important to 
acquire high dynamic range (HDR) images to recover all scene 
colours (Reinhard et al, 2005). 
2.4 Interactive Visualisation 
The ability to interact with 3D models is a continuing problem 
due to the fact that the demand for detailed model is growing at 
faster rate than computer hardware advances. The rendering 
algorithm should be capable of delivering images at real-time 
frame rates of at least 20 frames-per-second even at the full 
resolution of both geometry and texture. We use the Atelier 3D 
system, a view-dependent real-time system for multi-resolution 
models. When at close up the full resolution is shown then it 
decreases when moving away. It is based on the GoLD system 
(Borgeat et al., 2007) described in section 1.3 above. 
3. MODELLING FROM RANGE SENSORS 
The steps for creating 3D models from laser scanning are well 
established (Bemardini and Rushmeier, 2002). Here, we 
summarise the acquisition, processing, and texture mapping of 
data from such sensors as implemented in this project. 
3.1 Field Work and Data Collection 
As with this type of project, adequate planning before the actual 
field work demands a systematic approach to identify the 
proper sensor technology, estimate time for different scanning 
methodologies, define quality parameters, etc. The fieldwork 
must be completed within a specific time dictated by the 
availability of equipment and support personnel, allowed access 
to the site, and project budget. Thus, it is important to assemble 
an optimum team on the site to handle all operations effectively. 
Five days with three persons were spent as follow: one person 
for scanning; one person for initial scan alignment (see section 
3.2), data backups, and general guidance; and one person for 
digital imaging for texture mapping and IBM. 
To satisfy the project requirement, the Surphaser & 25HSX TOF 
phase-shift based laser scanner was selected (Figure 5). It can 
acquire the data at about 5 m range with a noise level of 0.25 
mm (standard deviation), and accuracy of less than 1 mm 
(maximum error). This has been verified with our own tests on 
and off site in our 3D metrology lab (Beraldin et al., 2007). 
Figure 6 shows the results of a test with the Surphaser* scanner 
that confirms its ability to capture sub-millimetre details. Other 
tested scanners failed to capture those details. However, to 
achieve this accuracy on marble apparent laser penetration 
errors (about 5 mm) must be corrected. 
Figure 5. The Surphaser® 25HSX laser scanner 
Figure 6. Validation test with the Surphaser* 
Figure 7. Area captured by Leica & HD3000 laser scanner