yrmation. 
ameters of 
to image 
/e the high 
trol points 
as well as 
> algorithm 
asis of the 
vo images 
cient. The 
ings of the 
je area is 
icient. The 
the nodes 
a priori. 
mid) about 
his we can 
erated by 
‚T,Z 
the X,Y,Z 
on is very 
s of the 
je use the 
n of X, Y,Z 
(31) 
left photo 
s of the 
and right 
2 system 
tain first 
ula of d, 
he ground 
  
Figure 3. The errors of the DTM. 
dc, dx, 2 dx, ; 
z 
b 
zat )db, + 
b^A c,bx C, p p 
  
  
d 2 ZU dis 5, db, +( 
eae EE Y 
2 
Yao (554599 4 L dre) + d 
2e 
cP pCt p 
(32) 
where A is the scale denominator. (Lobanow, 1984) 
We take into account that in differentials dx] and dx; are 
included the errors of the "absolute orientation and in the 
dp’ we consider only influences of the elements of the 
interior orientation. 
Figure 3 shows the distribution of the Z coordinate error. 
The grey levels coorespond with the error from 0,4m to 
0.5m. 
5 ERRORS OF ORTHOIMAGE 
The correctness of obtained image depends not only on 
errors of the calculation of grey levels in the resampling 
process. We should, first of all, take into account all 
variables in transformation which was used in the 
generation of the orthoimage. Looking at the formula (31) 
and (14) we can see that we should analyse errors of 
the determination of the point position in the photo which 
results from the errors of the parameter estimation of the 
exterior orientation and appropriate determination of the 
Z coordinate from DTM. 
After we had made the relative and absolute orientations 
of stereo pair we received the project transformation 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
introducing the formulas of model coordinates 
Rp - Rp * ApA(R-Rg) (33) 
into formulas 
X Y 
a es yo o, (34) 
where 
R, =[X,.Y,.Z,]' the model ccordinates of the point 
o 0 X0 947 i 
R? z[X9, Y5, 22] the translation vector 
A the scale denominator 
A rotation matrix 
R-[X,Y,Z] the ground coordinates of the point 
R, =[X,,Y,,Z,] the mean vector 
The mean square errors of the image coordinates m,, m, 
we receive calculating differentials of (34). We assume 
(as usual) that rotations angles are small (Wang, 1990). 
The results of the calculation of the error m= Im? «m; 
are shown in Figure. 4. 
The grey levels correspond with the error from 0.15mm 
(white) to 0.21mm (black) when the image pixel size was 
0.06mm. We can see that this errors depend on the 
distant from R,. The influence of the errors of the 
estimation RP is big and constant in the whole area. 
The effect of the error of the point position in the photo 
will influence on the error of the grey level determination 
of the pixel in the orthoimage. 
Finally using the function (4) we have 
=a’m? UM (35) 
2 
Mey) 
where a, ,a are the values of the partial differentials of 
the function (4) at the point (x,y). 
6 CONCLUSIONS 
We want to inform that all tests we performed using our 
programme on the PC computer. We have written it in 
Borland C++ ver. 4.5 language in Windows 95 
environment. Programme is working with Paradox 
Database for Windows. The coordinates of control points 
are inserted into tables of this database. The data 
computed during the DTM generation process by 
correlation method are stored in special files. Figure 3 
shows that not all parts of the area have Z coordinates so 
the needed additional measurements are stored in