TM 
  
  
(65) 
and correspondingly from the equations (3) and (4) 
= =} AT | 
Qpp = Aon t A25 Ajo Qui Ai2 A22 (66). 
The computation of the matrices Qj or Qoo can not be performed by stepwise 
accumulation and therefore & considerable computing effort becomes necessary. 
On the other hand, the relative accuracies within a model are often of pro- 
nounced interest. This fact underlines the advantages of & solution which ig 
established either directly on formula (32) or on the reduced normal equation 
system (38), provided that the corresponding computations can be handled by 
electronic computers. If it suffices to obtain only the mean errors of the 
eliminated parameter corrections, it is possible to compute the squares of the 
corresponding weighting coefficients as the diagonal terms in (65) or (66) by 
‘a stepwise accumulation in the same way as the reduced normal equation system 
was stepwise accumulated. 
In case the elements of orientation can be assumed as flawless, the corre- 
sponding Qoo matrix becomes a null matrix and Ai matrix emerges as the 
weight matrix of the triangulated points. The more excess observations in- 
corporated into the original least squares solution for the orientation 
parameters, the better this approximation solution will be. 
G. An Example of the Described Solution Using Functional Schematics 
Figure 5 shows the overlap of 3 photographs which may be considered as 
being taken either by aerial or ground established cameras. The different 
types of control points. are marked by the following symbols: 
absolute control point, given by X, Y, 2 
partial control point, given by X and Y 
partial control point, given by Z 
relative point 
hl 
rpm