CIPA 2003 XIX 11 ' International Symposium, 30 September - 04 October, 2003, Antalya, Turkey 
520 
(a) left image (b) right image 
Figure 6. Stereopair of the upper part along the diagonal direction of the statue base 
targets would be in varying distances from the camera, 
following in effect the relief of the sculpture. For each 
stereopair, different locations were selected for the sticks. The 
sticks were fixed stable on the scaffolding levels of the wooden 
framework and the XYZ coordinates of each control point were 
measured using a total station with an accuracy of 4mm. In 
addition, the coordinates of natural points on the surface of the 
sculpture, that were selected very carefully to be used as control 
points, were measured. Yet, very few of them were used for the 
orientations of the stereopairs due to the difficulty of their 
positioning with adequate accuracy and reliability on the 
photos. Figures 3 and 6 show examples of the predefined 
control points, while Figure 6 shows seven selected natural 
control points marked in red (small circle) on the digital image. 
3.3 Laser Scanner Data Collection 
The laser data capture was performed by Archaeoptics Ltd. 
using a Minolta VI-900 laser scanner (http://www.minolta- 
3d.com), which is a laser light-stripe triangulation ranger- 
finder. The scanner provides colour data with a CCD resolution 
of 640x480 pixel per colour and has three interchangeable 
lenses of f = 25.5mm, 14.5mm and 8.0mm respectively. The 
highest resolution the scanner can provide is 170pm and its 
precision is ±0.008mm. The scanning procedure is safe for the 
sculpture, since the 690nm red semiconductor laser beam 
moves continuously during scanning for extremely small 
duration. After setting up the scan, the actual process takes 
about 3 seconds to acquire roughly 300,000 points. 
The scanning procedure requires an operator who interactively 
moves the scan head to set each new scanning window. The 
window is constrained by both the field of view of the lens 
being currently used and any occlusions of the laser or the 
camera mounted on the front of the scanner. The scanner has 
the ability to acquire RGB data which was used only for 
determining material changes and breaklines in the statue. The 
duration of the Hermes scanning was 2 days during which 649 
overlapping scans (20-30% overlap of adjacent scans) were 
acquired at a distance of about lm from the statue and at 0.5mm 
resolution with an accuracy of 0.25mm, in order to record 
details smaller than 1mm without the dataset becoming 
completely unmanageable in size. Each scan comprised the 
scan metadata, that is laser power used, focal length etc, the 
raw 3D coordinates of the range grid, intensity and per-vertex 
colour information. In total more than 269 million triangles 
were created. The raw data required around 10Gb of space 
uncompressed (Tsakiri et al, 2003). 
4. DATA PROCESSING 
The data derived from both methods were edited independently, 
so that a 3D model as complete as possible would be acquired 
from each one. At a second stage, the data from both methods 
will be used in an integrated mode so that the best result would 
be achieved. 
4.1 Photograminetric Procedure 
The process described in this paper refers to photogrammetric 
images derived from the digital camera. The analogue images 
are processed independently and a comparison considering the 
accuracy and the quality of the results from each type of image 
is the focus of a future study. 
The first stage in processing refers to the calibration of the 
SONY digital camera for taking photos from a distance of 
1.80m with the lens at the minimum zoom. A 3D close-range 
test field of 6x6x3 m 3 was used, which is established at the 
metrology room of the School of Rural and Surveying 
Engineering of NTUA and is measured with an accuracy of 
2mm. The test field consists of approximately 130 targets in 
total, placed on stable metallic columns and a metallic board 
attached onto a wall. Part of this test field was photographed 
from six different positions and orientation angles. A 
simultaneous adjustment was made for the data of each photo, 
the exterior orientation and the focal length, principal points’ 
coordinates and the two coefficients of the radial distortion, by 
measuring 22-30 control points, through in-house processing 
software (Toumas, 2003). The results with their respective 
standard deviations, which are notably low, were estimated as: 
c = 2 913 ± 4 pixel, 
x 0 = - 22 ± 6 pixel, y 0 = 3 ± 2 pixel.