42 BALLISTIC PHOTOGRAMMETRY, SCHMID 
tical photography flown with 66% overlap for 42 photographs featuring two absolute 
control points at the beginning and the end of the strip, respectively. Figure 2 shows the 
accuracy with which the elements of exterior orientation are cbtained in the center of 
such a strip for strip lengths varying from 5 to 54 photographs. 
Equally important are the non-diagonal terms in the inverted normal equation sys- 
tem representing the correlation numbers. They display the correlation existing between 
  
  
  
  
  
  
   
  
Áo man * Ka pe 
fries" > Kap | 
my! s Ke fe J 
p mean square error of unit weight 
in microns 
(for Fig land 2 ) 
3.0 
Dy, Un] : Ky M 107 5$ 
my, £m] : Ky, M-107$. M (ferfig.land 2 ) 
mz, [Im] : K, e M- 1075 
     
p: mean square error of unit weight inmicrons 
M-H (scalefactor) 
  
— . * 
5 10 15 20 25 30 35 he 45 50 ss Vo.ofphotographs in the strip 
Fig. 2. 
the various orientation parameters and the coordinates of the model. Both the propaga- 
tion and the correlation numbers are not only significant for error theoretical studies, 
but they provide also an excellent means for determining objectively, manufacturing 
tolerances for the design of restitution equipment. 
The significance of the results obtained with the propagation and correlation num- 
bers depends on all regular or systematic errors being sufficiently small, so that they 
are submerged in the noise level caused by irregular or accidental errors. 
The fundamental observations used in the analytical treatment of photogrammetric 
data are image coordinate measurements. Two distinctly different sources of error con- 
tribute to the residual errors of these basic measurements, namely, the group of errors 
associated with the measuring process on the comparator, and the group of errors origi- 
nating from the process of taking the photographs. 
The principal residual errors in the coordinate measuring process are setting and 
reading errors; it is typical for these errors to have a normal distribution. In addition, 
there is a host of small errors, which remain from the necessarily imperfect comparator 
adjustment and calibration. The central limit theorem supports the assumption that the 
sum of a great number of such residual errors is also normally distributed. 
Skillfully manufactured grids can provide the equivalence of idealized photography. 
It is then a straightforward matter to determine the mean square error of the measuring 
process, associated with a specific coordinate measuring device, provided the grids have 
been calibrated to an adequate degree of precision.