sults 
r the 
  
reci- 
Was 
cies. 
]ved 
sible 
iccu- 
5mm 
data 
low. 
3ated 
] that 
n the 
ld is 
l(see 
; Was 
rtical 
re are 
arger 
offset 
eight 
   
station film-based calibration networks were taken of 
the test field; one before, and one immediately after the 
video portion of the test. 
  
igure 
Re-furbished Test Field 
  
  
  
A set of 120 video images was collected and measured 
for this test. The 120 images were taken in two steps . 
The first 60 images were collected by setting the 
camera roll to 0° and by rotating the test field in 6° 
increments on its rotating base. Similarly, the second set 
of 60 images were collected with a camera roll of 180°. 
Data analysis was performed using 8, 16,30,60and 120 
station network sets of the collected data. Target im- 
ages occupied approximately 100-150 pixels on the 
Sensor. 
Target images were driven to automatically using a 
curve prediction algorithm. This required an operator to 
initially identify targets on three frames by using a 
pointing device and a graphics monitor. The remaining 
frames were measured automatically at a rate of about 
15 seconds per frame (including rotating the test field). 
No prior knowledge of the object was required. 
The results are divided into two sections. The first set of 
results is from five different combinations of runs from 
the set of 120 images measured. Analysis was per- 
formed on 8, 16, 30, 60 & 120 station sets of this data. 
The second set of results is a repeatability test composed 
of portion of the 120 image set broken into eight 
independent networks of 8 stations each. 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Closures Results for Videk Camera (microns) 
# of Stations | S s. Sad eomm 
8 28 25 27 1/25 
16 28 25 27 1/25 
30 28 23 26 1/26 
60 23 19 21 1/32 
120 37 23 25 1/27 
Table 5 
  
  
  
The first set of results from this section is the RMS of 
  
   
   
  
  
  
  
  
  
  
   
   
  
  
  
  
   
  
  
  
   
  
  
   
  
   
  
  
  
  
   
  
  
  
   
    
   
   
   
   
   
   
   
  
   
   
   
   
    
    
   
      
closures for the bundle adjustments for each of the run 
combinations (Table 5). 
Although these closures are consistent with previous 
measurements of target images of approximately 150 
pixels, pointing precision on the order of 1/50 pixel or 
better for images of this size is expected. However the 
thresholding technique suggested by (Trinder 1989) has 
not yet been incorporated in the measurement process. 
This may explain some systematic effects in the results. 
For 256 grey levels Trinder suggests: 
T=74.(SF)!*A-"! 
where SF refers to the 2-sigma-width Gaussian 
Spread function and A the target size 
The second set of results is the estimated and actual 
accuracy summary for each combination of runs (Table 
6). 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Estimated and Actual Accuracies for 
Five Combinations of Runs (mm) 
Estimated Actual 
8 0.025 0.023 
16 0.017 0.018 
30 0.011 0.017 
60 0.008 0.016 
120 0.006 0.018 
Table 6 
  
  
  
As expected, the accuracy estimates improve propor- 
tional to the square root of the number of stations. The 
actual accuracies, however, show no significant im- 
provement after 16 stations. This is not what would be 
expected if the errors in the measurement were pre- 
dominately random. It is therefore believed that 
unmodeled systematic error remain in the data. Figure 
5 shows a graph of both estimated and actual accuracies 
asafunction ofthe number of stations. Also shownisthe 
theoretical curve of accuracy improvement. 
The graph does illustrate to a certain level that the 
addition of multiple stations improves accuracy. In this 
case accuracies improved from 1:50,000 in the 8 
Station set to accuracies in excess of 1:70,000. 
The second section of this test is a repeatability test 
made by separating the 120 video images into eight 
independent data sets. Each of the data sets was reduced