Since 
ie two 
tem is 
inates 
rsors. 
otions 
kward 
ys the 
Lipton 
tively, 
ystem 
licates 
e two 
button 
S way 
/ most 
ftware 
r. The 
initially 
e, the 
"CPU. 
ng the 
ement 
'ertical 
image 
linates 
e area 
osition 
rsor is 
ding to 
(7) 
on and 
sors in 
e. The 
olution 
BLANK 
(8) 
ited by 
ditional 
mages, 
nts to 
of this 
‚a 1GB 
120 Hz 
talEyes 
a major 
ve is to 
juakes. 
cultural 
rone to 
For the above purpose models of ancient monument parts are 
subjected to seismic tests using the 6DOF earthquake 
simulator of the Laboratory of Earthquake Engineering of 
NTUA. Since the movement in space of the test objects would 
be rather unpredictable continuous monitoring was required. 
Hence stereoscopic video was employed and the system 
described has been developed. The details and the first results 
of these experiments have been presented elsewhere 
(Georgopoulos et al. 1995). A brief description will be given 
here in order to present the applicability of the system. 
The initial tests involved research into all aspects of the 
experiments. Namely the video cameras to be used, the 
construction of the whole setup and the algorithms employed 
in order to achieve the best possible results. The first 
simulation experiment was taped using two commercial 
camcorders. One was a SONY Handycam 10x Video 8 AF 
with 310.000 pixels and the other a SONY Handycam Pro 
Video 8 AF (V90) with 440.000 pixels. Around the object and 
independently thereof a test field was setup with 16 premarked 
targets. The targets were black circles on white background, 
approximately 30 mm in diameter. In addition three targets 
were attached on the object itself, in order to enable the 
determination of its displacements. The co-ordinates of the 
targets were determined using a Leica T1010 electronic 
theodolite equipped with a Leica DIOR 3002S EDM, with an 
accuracy of £3 mm. The two cameras were positioned on 
tripods at a distance of 5 m from the setup with a base of 
0.70 m, thus providing a base-to-distance ratio of 1:7. Imagery 
was acquired simultaneously with both cameras and was 
recorded on VHS 8mm tapes. The frames were grabbed using 
the Screen Machine Il from FAST Electronics frame grabber 
with a resolution of 736(H)x560(V) pixels. This commercially 
available frame grabber has the usual standard features and is 
escorted with an image editing software with rather limited 
capabilities. 
In order to overcome certain algorithmic problems involving 
two different cameras into the calculations, a second series of 
experiments was conducted with two identical Panasonic K900 
VHS video cameras. The same targets were used and they 
were measured with greater accuracy using only electronic 
theodolite measurements from two stations. Their co-ordinates 
were determined with the help of intersection in space with a 
final accuracy of +1 mm. In another experiment (Figure 3), 
where a 1:3 model of a Parthenon pillar consisting of eleven 
cylindrical rings was tested two identical professional Beta 
video cameras were used. These cameras were offered by a 
major Greek commercial TV Channel and were genlocked 
specifically for this experiment in order to produce absolutely 
synchronised frames. They were able to record 50 frames per 
second, thus increasing the sampling rate of the moving object 
and the amount of the data stored. 
Several algorithms were either developed or used from 
previous work. These algorithms involved mainly a specially 
developed target location algorithm, a suitable camera 
calibration procedure and the necessary calculations for 
determining the ground co-ordinates of the points of interest. 
There were three presignalised points on each pillar ring. 
By calculating the co-ordinates of these targets in space the 
absolute position of each one of the object blocks would be 
determined. This determination was carried out both 
monoscopically, for the case where the object was forced to 
Swing in one plane, and stereoscopically. 
In cases the pillar would perform displacements in 3D space, 
the object co-ordinates of the observed points are calculated 
by a two camera triangulation algorithm. Image co-ordinate 
measurements of the points of interest are performed 
115 
automatically on simultaneous views of the object in order to 
avoid manual identification of the targets on each frame. 
Simultaneity is achieved by matching the frames with the same 
time clock indication. Since absolute synchronisation of the two 
cameras in the initial experiments required special hardware, 
which was not available, the synchronised frames were 
manually determined. Although the proposed method is very 
simple, many problems may occur in practice. The most 
important problem is that the clock is not clearly recorded in 
all video frames. Hence, due to image deterioration in certain 
cases it was necessary to pick frame pairs with frames having 
a time difference of as much as 0.05 sec. This was, however, 
considered as having little effect in the final result. Of course 
there was no synchronisation problem the case of the 
experiment taped with the two genlocked cameras. 
The image co-ordinates of the observed pillar points are 
introduced into the computation as tie points with unknown 
positions. For each tie point 3 additional unknowns and 4 
observation equations are added. 
A RMS error from the comparison between the object points' 
co-ordinates and the co-ordinates computed by the 
triangulation was calculated. Concerning the first experiment 
with the two SONY cameras, the observed errors were 4 mm 
in the X and Y directions and 8 mm in the Z direction. The 
results are better in the case of the two Panasonic KS00 
cameras, where the observed errors were 1.5 mm in the X 
and Y directions and 4 mm in the Z direction. 
These results are quite encouraging considering the hardware 
  
  
  
-6 
-8 : ui 
0.00 2.00 4.00 sec 
wa a wl UDECK 
—— CONV. METHOD 
———UWMIDEO 
Figure 4 
used. Moreover the amelioration of the final accuracies in the 
case of the second experiment is not a result of the different 
cameras used, but of the better accuracy of the measured 
control points’ co-ordinates. 
After the calculation of the co-ordinates of the pillar points for 
the whole duration of the experiments a comparison was 
attempted in order to assess the relative and absolute 
precision of the method. The comparison was performed with 
the similar results obtained from the conventional displacement 
meters measurements of the same experiment. Moreover the 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996