inthe case of the maximums, estimated accuracies were 
3 times the required level. A notable exception was in 
the axis aligned with the axis of the reflector. This axis, 
which is the most important, had estimated accuracies 
which all were within the requirements, even the maxi- 
mums. By virtue of these results, it was deemed 
appropriate to investigate further. 
A full reflector survey was not practical initially so a 
simulation of a portion of the reflector was suggested. 
A 3 camera segment of the 24 camera network observ- 
ing a triangular “petal” of the reflector was simulated 
(see Figure 3). Because the 3 station network is not as 
strong as the 24 station network, the expected results 
were not as favorable. As long asthey were predictable, 
this situation was acceptable. For this reduced network, 
  
TC) 
dq 
Figure 3 
3 Camera Station Geometry 
  
  
  
computer simulations predicted accuracies 3 to 4 times 
worse, especially in the axial direction of the reflector. 
In order to simulate this measurement in a more practi- 
calandtimely manner, in preparation for a measurement 
on the actual dish, a 4 scale version of the network was 
created at GSI in Florida. This scaled version was to 
serve as a proof of concept. The !/4 scaling was applied 
to the size of the object, the size of the targets, the 
positioning of the camera, and the expectations for 
accuracy. For this test, a single video camera was used 
and moved into the three positions required. The 
simulated object was very stable (concrete), so no 
deformation was experienced during data acquisition. 
Targets of about 12mm were used, resulting in images 
containing approximately 50 -100 pixels. Concurrent 
to the collection of video information, a film-based 
network was taken using a CRC-2 to calibrate the target 
     
  
  
  
  
   
    
  
  
  
  
  
  
  
   
  
  
   
  
  
  
  
  
  
  
  
  
   
   
  
   
  
  
  
  
  
   
  
  
   
  
  
  
   
  
   
  
  
   
    
field. This allowed a check of the actual accuracy of the 
video-based measurement as the film-based results 
were more than 5 times more precise. 
A summary of the predicted and actual errors for the 
video measurements are listed in Table3 below. 
  
Summery of Accuracy Estimates (mm) 
  
Simulated Actual 
  
X mm | Ymm | Zmm | Xmm | Ymm | Z mm 
  
RMS is 0.203 | 0.163 | 0.097 | 0.465 | 0.381 | 0.208 
  
Minimum is | 0.091 | 0.025 | 0.076 | 0.211 | 0.061 | 0.165 
  
Maximum is | 0.391 | 0.264 | 0.119 | 0.914 | 0.648 | 0.254 
  
  
  
  
  
  
  
  
  
Table 3 
The computer simulation assumed measurement preci- 
sion of 1/50th of a pixel and only about 1/25th was 
achieved, thus the difference in estimated accuracies. 
This is most likely because of possible unresolved 
systematic errors in pointing precision and possible 
instabilities in the lens elements. 
The film-based measurement had an actual RMS accu- 
racy estimate of about 0.075mm in plane, and 0.038mm 
out of plane. The comparison of the video-based data 
and the film-based results are listed in Table 4 below. 
  
  
  
  
  
  
  
  
Transformation results Video vs. 
Film (mm) 
vX Vy Vz 
0.434 0.234 0.198 
Table 4 
These results are consistent with the actual estimated 
accuracies. As a result of this test, it is expected that 
mid-1992 a test using 3 separate video cameras on the 
actual reflector will be carried out by GSI. 
3.3 TEST NETWORK 3 
The third test took place in 1992 and the test field is 
essentially the same asused in TEST NETWORK 1(see 
Figure 4). In this new version the grid of targets was 
replaced by “fresh” 12.7mm retro dots and the vertical 
offset targets now occupy points on the grid. There are 
no 6mm targets placed on the new field as the larger 
targets produce superior results . Six targets were offset 
from the plane by 10mm to 25mm shims. Two eight