RMS error [ 
bo 
  
  
  
5 10 I 120 
Number of selected points 
Do 
WI 
Figure 2. Selecting a varying number of tie points in each tie 
point area. RMS errors in X, Y and Z in the check points, as a 
function of the number of selected points. 
3.1.2 Test material 
The OEEPE test block Forssa was used in the investigation. The 
block consists of 28 digital monochromatic aerial images in the 
scale 1:4000 (four strips with seven images in each, 60% 
forward overlap, 20-40% side overlap, 30 jim pixel size). The 
images are from an urban area. The control point configuration 
(14 XYZ) and the check points (50 XY and 41 Z points) were 
the same as used in the OEEPE test project, see (Jaakkola and 
Sarjakoski 1994). 
A large number of tie points was measured automatically by the 
system described in Section 2. The same data was used 
throughout the investigation, with additional measurements in 
some cases. Also interactive tie point measurements were 
carried out, for comparison. The fiducial and ground control 
point measurements were carried out interactively. 
In the automatic tie point measurement, 102 tie point areas were 
selected in the standard Gruber positions. About 80 tie point 
observations were measured in each of them (in the image 
pyramid layers, respectively 4, 4 and 5 tie points). 
The RMS errors in the check points, when using automatically 
measured tie points, were: X: 2.3 cm, Y: 3.0 cm and Z: 3.5 cm. 
When using interactively measured tie points, the RMS errors 
were: X,Y: 2.3 cm and Z: 3.3 cm. 
When using automatically measured tie points, the RMS error 
in Y was considerably worse than in X, with a difference of 
almost 1 cm. This phenomenon was not present when using 
interactively measured observations, so the reason is likely to be 
the existence of poor observations. 
3.2 Results and discussion 
3.2.1 Selecting a varying number of points from each tie 
point area 
A varying number of points (2, 3, 5, 10, 15, 20, 25 and 30) was 
selected from the test data in each tie point area in each image. 
In the selection process, the criteria presented in Section 2.1.3 
were used. 
340 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
  
  
  
  
  
  
z^ "Geile e viso enge enr 993 1 voles = 
m, 
Eam um mr urn bee bene 
e 
ET emrmraasl 
= 
ii nn. ss 
0 | f 
2 5 10 15 
Number of selected points 
b) 
S 
€ 
= 
S 
© 
un 
= 
"A A A 1 Le 11112120 
0 + 
  
  
  
  
RMS error [em] 
  
  
  
Number of selected points 
—€— 3x3 —ü— 5x5 
Figure 3. Using 3x3 and 5x5 tie point area distributions. RMS 
errors in the check points as a function of the number of 
selected points in each tie point area. 
Some comments should be made about the selection process. 
The distances between the selected points decreased as the 
selection process proceeded (the minimum distance between 
points was 12.5 pixels). In general the required number of 
Observations was found, but not in difficult cases (which seldom 
occurred). The completeness of the observations was good 
when up to 10 points were selected, thereafter it was slowly 
decreasing. Due to this, the significance of the selected points 
will probably decrease as the selection process proceeds. 
The RMS errors in the check points are presented in Fig. 2. The 
following conclusions can be drawn: 
e The number of points had no influence on the accuracy of X 
and Y. On the other hand, the number of extracted points 
   
   
   
  
  
  
  
   
  
  
  
  
  
  
    
  
  
  
  
  
   
    
   
  
   
   
   
  
  
   
  
   
  
  
  
  
   
   
   
   
    
RMS error [cm 
N 
[cm] 
L3. 
ITO 
9 3 
MS 
X 1 
RMS error [cm] 
Qm Bop Usb um 
Figure 
eITOIS 
selecte 
spl4: 
splittin 
splittir 
RMS ¢ 
in the | 
ha 
to 
e It 
op 
cn 
nu 
3.2.2 
A 5x5 
effect