model matrix. The mutual relation between 
them is of no correlation (expressed by nulls). 
The conditions matrix (B) and the weight ma- 
trix (P) are defined as: 
  
xml] 
Vyml 
Vzml 
* 
ARS a ^ | 
Be 0 ka * " 
xgl 
0 ( 3nx3n 
Vygl 
Vzgl 
P. S rey, od EF 0 b 
0 Yos ground 
  
The solution provides the  variance- 
covariance matrix of the computed parameters, 
that is expressed as: 
Y4700*N^ 
where: 
Nd EA cie rc. 
n-u 
3. DIRECT MEASUREMENTS 
The next step that follows the transfor- 
mations’ error determination, concludes the 
error's propagation and formulate a mechanism 
that generates the measurements SD. By gen- 
erating the mechanism, the accuracy evaluation 
of the spatial data (the observations them- 
selves) and of it's derived applications, is en- 
abled. 
3.1 Variance and covariance ground coor- 
dinates 
The ground coordinates' accuracy is gen- 
erated, again, by the variance-covariance 
propagation technique. There are two elements 
that effects the coordinates accuracy. The first 
is the absolute orientation transformation er- 
12 
rors, and the second is the model coordinates’ 
errors (caused by the relative orientation 
transformation error). The transformation on 
which the variance-covariance propagation is 
applied, is the projective transformation. 
A variance-covariance propagation expressing 
those two transformation errors is defined as: 
S nd = H * S MET * F, * MET 
where 
F, - Elements referring to the absolute 
orientation parameters. 
FE, - Elements referring to the model 
coordinates. 
The model coordinates component con- 
tributes a non negligible value to the computed 
variance. For example, model coordinates de- 
rived from photographs at the scale of 
1:40,000, contributes about +1.5m to the co- 
ordinates' accuracy. Therefore, as the control 
points are more accurate, the model coordi- 
nates component is more effective. 
The following sub-chapters describe two 
applications of error propagation implementa- 
tion. 
3.2 Equi-error contours 
Accuracy of data derived from photo- 
grammetric measurements, is usually evaluated 
by the absolute orientation’ SD and is consid- 
ered as fixed value for the whole stereo-model. 
Practically the accuracy varies through out the 
stereo-model in a manner that measurements’ 
SD at the center of the model are smaller than 
those collected near the edges. The error 
propagation mechanism, as formulated above, 
enables a precise determination of the meas- 
urements accuracy, as well as evaluation of er- 
ror variation. 
The equi-error contours, illustrate the SD 
through out the model by presenting them as 
contours. The provided map enables a more 
accurate S 
use of a fix 
of a stere 
based on g 
cally pass | 
lustrates a 
(errors dep 
  
  
Fig. 2. Sm: 
The 1 
first refers 
by which it 
the second 
from the c 
contours in 
reo-model : 
+3 m at th 
edges, a ra 
glected. 
The € 
the use of 
set of meas 
(for each r 
vant contol 
accuracy re 
tours are de 
grid (althou 
data set for 
3.3 Profiles 
À pr 
another pı 
mechanism, 
results eval 
sential tool 
sis, is defin 
line, that at