Natalia Moskal 
  
The same situation is noticed for correction directions (horizontal and vertical angles) which may be presented as space 
coordinates functions of the center of projections and the point of the object. Thus this type of equation is reduced ty 
the model with the correction Y po» an absolute term Y B and matrixes of partial derivatives B and Ds : 
System (4.1) presented in general terms will look as follows: 
€ = Ÿ —- AX - DZ, (42 
y =Y— TX - FZ. 
Vectors X and Z are to be found due to certain priorly defined conditions, that characterize stochastic nature of 
probabilistic model. This conditions may be as follows: 
Derrors € and Y are influenced by the Gaussian law of division. They are mutually uncorrelated and uncorrelate( 
inside groups € and 'y (this is classical method of the smallest square; in this case the diagonal matrix of weighs i; 
known for vectors £ and y : 
P = Pe ; (43) 
Py 
2. errors £ and Ÿ are influenced by the division law which is different from Gaussian. They are mutually uncorrelated 
but correlated inside groups € and "y , thus co-variant matrix of measurement errors is known — 
Eo. à 
The correction minimization model is adopted for errors with Gauss division and matrix of weighs (4.3) 
£g p.£*y P, 17 min (45 
And for errors with division different from Gaussian and matrix (4.3) we may apply the condition of minimization d 
unsquare function of loses, or the condition of minimization of mixed function of loses [5]: 
2+d : 
= min (46 
  
e ly 
Where d — parameter of square - — <d <0 
Model (4.2) with condition (4.6) and covariant matrix (4.4) is the most general from theoretical view. 
Let us show (4.2) in the following way 
IEEE AE] à 
or T|  R- SU 
  
632 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
Accordi 
get adju 
In the rec 
The valu: 
Where n - 
T — 
In mixed