coefficient of two test statistics, we can divide the Separa- 
bility into three classesaccording to the maximum correlation 
coefficient: 
--- good separability, L£ (9,1) max 75%, 
--- bad separability, if 75% < (oi ax £ 100%, and 
--—= Not separable, if P1j is in any direction equal to 100%, 
Summarily, the steps for the analysis of reliability under two 
alternative hypotheses are namely 
(i) to give matrices A,P.and H, (ie 1,2); 
(ii) to calculate the matrices Qo (75631; ;M45 and the cor- 
relation coefficienfs 014» 9G» ® max? 
(iii) to estimate the internal and external reliability and 
separability values according to Eq.(19-29), ana 
(iv) to evaluate the reliability of the given geometry and 
to optimize it in consideration of accuracy, reliability and 
economy, 8 
3. SOME APPLICATIONS 
3.1 lod ility of gross errors in relative orientation 
The lower bounds of detectable and locatable errors are calcu- 
lated for' both cases of analytical relative orientations with 
and without scale transfer respectively. The results are showed 
in Fig.4 and the relationship between the maximum lower bound 
values and the point distributions are given in Fig. 5a and 5b. 
The conclusions are 
=== Only using six Gruber points the gross errors are very 
difficult to be detected and can not be located. 
--- The gross error detection is almost. independent on the 
method of relative orientation if with or without scale trans- 
fer, while 8ross error location is dependent on that. The re- 
lative orientation with scale transfer will lead to the better 
locability of gross errors. This is the reason why it is usually 
used in the on-line aerotriangulation. 
--- The use of group points is useful for the detection of 
gross errors, while a uniform distribution of tie points is 
available for the location of gross errors, 
- 194 -