hul 2004 
d digital 
ls) fully 
r set up 
ies. This 
Vere not 
oud data 
P1-3000. 
000 and 
polygon 
h global 
re used. 
indicate 
hown on 
  
tographs 
'hurch II 
pictures 
e length 
nm. The 
; mm for 
cialist in 
tographs 
1mber of 
s 2 mm. 
he small 
vall (see 
he actual 
Figure 9 
pe of the 
io model 
ventional 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 
  
  
  
  
m1 ; 
3n To ER 4 
= 
TRE A mie 
  
  
Figure 8. 3D model produced from two pictures using PI-3000: 
with texture (above) and without texture (below) 
  
  
Figure 9. Different view of the 3D model produced from two 
images 
To evaluate its accuracy, the 3D model was compared with 
elevations drawn by members of the Research Group of 
Byzantine Lycia. It took 800 man-hours at the site to carry out 
measurements using a non-prism total station and levelling 
string, and 300 man-hours to draw seven elevations in the office. 
In contrast, our method required one day at the site to take 
photographs and measure reference points, and one day to 
analyze data and construct the 3D model in the office. The post- 
process time would be shorter, maybe one or two hours, if the 
digital photograph were taken based on the predefined set up 
condition since it took 30 minutes for stereo matching on a PC 
with a Pentium IV 2.66GHz CPU. When compared with the 
elevation, the model was accurate to within 1 cm as shown in 
Figure 10. The result was satisfactory in terms of efficiency and 
accuracy. 
  
Figure 10. 3D model of Church II with overlay of elevation 
4.5 Orthophotograph of Floor Mosaic 
There are still wide areas of mosaic on the floor of Church III. 
The mosaic is composed of tesserae, small pieces of colored 
stone that measure approximately 1 x 1 x 1 em. Since the LPM- 
25HA cannot scan the shape nor capture the color of the mosaic 
floor, we were unable to model the details of the floor (see 
Figure 4). As a result we attempted to use digital 
photogrammetry to make an orthophotograph and 3D models of 
the floor mosaic. 
Due to difficulty with scaffolding, we were not able to take 
pictures from just above the floor and hence the photographs 
were taken from a slanted angle. A total of 25 photographs were 
taken, which covered about a 5 x 1.5 m area of the floor mosaic. 
Four reference points, with coordinates measured using the 
RTK-GPS, were used. It took about two hours for stereo 
matching producing 700,000 points. Totally it took 3 to 4 days 
due to the difficulties in selecting stereo pairs from pictures 
taken from a slanted angle. An orthophoto created using PI- 
3000 is shown in Figure 11 (left); Figure 11 (right) shows the 
photo with an overlay of the drawing. The overlapping accuracy 
was within 1 cm. 
Whether or not the drawing is a benchmark is arguable since 
there is a possibility that the drawing contained large errors as a 
result of the recording method. Researchers put a vinyl sheet on 
the floor and traced the pattern of the mosaic. Back at the 
office, the vinyl sheet was then spread out on the ground and 
photographed from a rooftop so that the sheet would be covered 
by a single image. The photo was projected on to a wall on 
which a section paper was hanging and lines were drawn on it 
following the projected pattern of the mosaic. The drawings 
therefore have both lens and projection distortion. Taking this