he 
© © 
o o 
© 0 Figure 1: Regular image point screens 
  
@-... 3x3 distribution 
o o o o © Q2, 45x5 distribution 
SR 
For vertical photography and flat object this set is outlined to be even independ- 
ent on the exterior orientation elements. Under very general conditions (large 
rotation elements, space character of the object) the corresponding sets can be 
extended by some of the 6 coefficients which are highly correlating with the 
6 exterior orientation elements in the case of vertical photography and flat ob- 
ject to obtain a complete bivariate set (here the parameters of interior orienta- 
tion are also included). With the creation of the complete set one has to pay 
attention to the sequence of the rotation elements. For example; if « is chosen 
dependent on $ and w then one of the two linear terms which lead to image torsions 
has to be rejected a-priori. 
Because of the possibly high correlations the extended sets have to be applied 
very carefully. To protect against overparametrization, i.e. to avoid the intro- 
duction of remarkable instabilities into the linear system (1) sophisticated sta- 
tistical test methods must be used for additional parameter significance testing 
in any case (see Grün /10/, /11/ and also section Dl of this paper). One has to 
note that the 1-dimensional Student test leads to wrong conclusions concerning 
the acceptance or rejectionof individual additional parameters if dangerous corre- 
lations do occur. Thus a more comprehensive treatment of statistical hypotheses 
testing methods becomes necessary. 
C. Accuracy and reliability of close-range bundle systems 
Hitherto in photogrammetric close-range systems the accuracy was almost exclusively 
in the center of interest. In this connexion sometimes a lot of work was invested 
to derive "accuracy predictors" which should represent a kind of accuracy models 
for special, often appearing network arrangements. Nevertheless, among all accu- 
racy predictors the inverse of the normal equations of a bundle system is still 
the best one. It accommodates every changing network configuration and every mode] 
variation. So the design of a project should be connected with the determination 
of the corresponding inverse for sophisticated accuracy studies. This becomes im- 
portant more then ever, since the introduction of additional parameter sets com- 
plicates an empirical accuracy appraisal. Besides the accuracy of a system its 
reliability should become a main objective of future studies. 
The reliability of a model describes its sensitiveness with respect to model 
errors. If a highly developed self-calibrating bundle model is used we may re- 
strict our considerations on the problem of gross errors.