ibul 2004 
While 
ates little 
standard 
ed sigma 
a strong 
h higher 
1e X and 
a strong 
Y Axis] 
3E-09 
24326 
72838 
1849 
  
PG and 
igher on 
r degree 
of error. 
much of 
mpirical 
rst scan 
om each 
ented in 
XIS 
7E-10 
98813 
50271 
95749 
39405 
13 
  
52405 
ollowing 
showed 
ined an 
NPG X 
r degree 
j X axis 
peaked 
vas only 
libration 
m NPG 
ontinues 
n° he 
jylowing 
  
  
   
calibration, indicating that some of the error was systematic 
in nature. Consistently before and after calibration, the PG 
unit performed at precision levels approximately ten times 
better than the NPG unit. 
32 Results of Spatial Error Analysis 
A global error analysis provided an indication of the overall 
pattern of error and the errors were also plotted to identify 
any spatially influence on the error. A sigma surface for the 
ten scans from each scanner was generated and presented in 
Figures 1 and 2. 
   
stat AA 
0.1 0.3 0.5 07 0.9 1.1 
Sigma (pixels) 
Figure 1 — Spatially plotted Standard Deviation from NPG X 
axis (left) and NPG Y axis (right) errors 
The error sigma surfaces for the NPG scanner show strong 
linear spatial patterns for both X and Y. Within very short 
distances in the scan, a high variability of error is seen in 
both axes. 
  
= 
0.03 0.055 
Sigma (pixels) 
Figure 2 — Spatially plotted Standard Deviation from PG X 
axis (left) and PG Y axis (right) errors 
In contrast, the spatially plotted standard deviation values 
from the PG scanner show a largely random pattern. On 
interest is the difference in scale between Figure | and 
Figure 2. The PG scanner produces scans with error sigma 
values approximately ten percent of the magnitude of those 
from the NPG scanner. There does not appear to be a 
systematic pattern to the error, qualitatively indicating that 
the errors are random in nature. 
After applying the empirical calibration, another set of 
spatial sigma surfaces were produced. Figure 3 shows the 
effects of the calibration upon the NPG scanned imagery. 
Similar systematic patterns remain, indicating that the 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004 
pattern is not a simple linear one and is not consistent 
between captures from the NPG scanner. 
  
x 
O 0.4 0.8 1.2 “41.6 2 
Sigma (pixels) 
Figure 3 — Spatially plotted Standard Deviation from NPG X 
axis (left) and NPG Y axis (right) errors after empirical 
calibration 
After calibration of the PG imagery (Figure 4), the pattern 
remains very similar and largely unchanged. This implies 
that little similarity of error exists from one frame to the next 
for correction. Again, it is seen that the magnitude of errors 
is approximately one tenth of those from the NPG scanner. 
0.005 0.015 0.025 0.035 0.045 0.055 
Sigma (pixels) 
Figure 4 — Spatially plotted Standard Deviation from PG X 
axis (left) and PG Y axis (right) errors after empirical 
calibration 
4 SUMMARY 
Following analysis of the global and spatial error patterns, it 
is clearly not appropriate for NPG imagery to be used in 
photogrammetric mensuration. The summary of statistics 
provided information regarding the large magnitude of error 
variation as well as the large absolute range of errors. 
Deviations of one pixel or larger are routinely seen and these 
errors would propagate through the photogrammetric 
workflow. Individual error spikes were seen up to 3.5 pixels 
and would not only influence interior orientation but would 
also lead to unacceptably high RMSE values during aerial 
triangulation and negatively affect the accuracy of any 
photogrammetrically derived product. 
Attempts to calibrate the NPG data empirically only saw 
limited success with a modest reduction in the range of 
errors. Large, multi-pixel errors still remain, as evidenced 
by the high sigma values.