tion 
ite, 
"of 
cion 
are 
rior 
sas. 
| as 
nal 
J to 
ing 
tion 
AP’ 5 
tion 
only 
ting 
rder 
Equation (1) and (3) were applied in the geometric model and 
as can be seen in Fig. 1, only significant AP's were selected 
after correlaton analysis of all unknown variables and after 
Significance testing of AP's. 
introduce to AP's 
  
STEP 1 
  
self-calibration bundle 
adjustment by all AP's 
3 
STEP 2 
  
  
  
compute 1)trace for cofactor matrix 
of object by elimination of AP's 
2) differences between standard 
deviation and coordinates 
| 
  
  
STEP 3 
  
search for data information to AP's 
(standard deviation, effection of 
each image points,correlations bet- 
ween AP's and unknown parameters, 
Search for standard deviation of 
iteration|| objects to existence of AP's 
i 
  
  
STEP 4 
  
compare with boundary value and 
result of STEP 2 
! 
  
  
  
STEP 5 
  
detect on unnecessary AP's based 
on digital informations obtained 
STEP 2 and 3 
pe 
eliminate unnecessary 
AP's 
  
  
  
  
  
  
  
  
  
     
  
adjusted resulting 
values 
Fig. 1 determinability test of AP's 
2.2 Detection and Elimination of Gross Errors 
The conventional data snooping method is inefficient in cases 
where more than one gross error exist, for many iterations are 
needed for such cases. In this study the progressive data 
snooping method was applied which detects and eliminates gross 
errors in a continuous way. The results are updated as shown 
and during adjustment of gross errors, only observations with 
errors will be used together with the formerly adjusted value. 
The progressive data snooping method used in this study is as 
in Fig. 2. 
373 
  
(‘ota observation values y) 
i 
least square adjustment 
  
  
  
  
  
  
  
  
Do,I-1, N 
  
calculate calibration 
Statistics amount and 
non-detected gross 
errors 
! 
detection of mesurement 
values that its studented 
residuals are greater than 
reject values 
i 
calculate on MEAN, MAX correlation 
coefficients for cofactors of residual 
to detected observation values 
i 
detect max studentized 
iteration residual and analysis on 
internal correlation 
i 
max internal correlation coefficient < 0.8 
! 
elimination on gross errors 
i 
updating of a posteriori variance 
; 
continuous adjustment with (n-1)observations 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
elimination of all 
gross errors 
  
Fig. 2 flow of progressive data snooping method 
3. ANALYSIS OF OBSERVATIONS 
3.1 Characteristics of Satellite Imagery and Formation of 
Input Data 
The satellite imageries used in this study are images of the 
same area aquired from different orbits and preprocessed in 
level 1B and level 1AP (photogrammetric films) and 1A (digital 
format in CCT). All satellite data are in panchromatic mode 
and the left image is vertical (L5937') and the right image is 
oblique (R26910'). Stereo model is formed as in Fig. 3. 
The base to height ratio is about 0.57 and the time 
difference between the left image (87.11.29) and the right 
image (87.11.30) is about a day, which means small difference 
of sun view angle which gives favorable observation 
conditions.