International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
Number of 2 3 4 5 6 7 
invariants K 
  
Total pairs 233 293 977 957] "948 23$ 
number. a=0° 
Valid pairs 34 75 105 106 113 118 
number, o=0 
  
Total pairs 343 321 288 269 250 232 
number. a=30" 
Valid pairs I9- os 66 iq 0 9 
number, a=30 
  
Total pairs 339 >320- 276-2270 > 245-> 246 
number, a=45" 
Valid pairs 11 55 57 66 79 37 
number, a=45 
  
  
  
Table 8. Number of properly tied features depending on the 
number of invariants K and rotation angle a 
From the above, verification distance criterion based on 
analysis of matrix (5) is also invariant to rotation of images. 
Hence the total image matching procedure under consideration 
is rotation invariant. 
2.6 Mosaic creation 
To perform the total montage of frames, the following 
successive tasks should be solved: 
l. search of tie points for pairs of successive images for 
each row, 
montage of separate raws using tie points, 
3. search of tie points between successive rows of total 
frame montage, 
4. montage of total mosaic using tie points. 
N 
In steps 1, 2 the coordinate table of successive frames is 
prepared and row montage is performed (Figure 9). 
  
Figure 9. Example of row montage 
[n steps 3, 4 the coordinate table of successive rows is prepared 
and montage of total mosaic is performed (Figure 10). Total 
mosaic consists of 1584 frames, each of size 400 kb. Elapsed 
computing time for processor Pentium-IV, 2.0MHz, amounts to 
40 min. 
3. AUTOMATIC RELATIVE ORIENTATION OF 
AERIAL IMAGES 
The relational matching approach developed above for camera 
frames junction, successfully works in automation of relative 
orientation for various types of imagery. All steps of 
processing, stated above for video frames, can be repeated here 
and supplemented with the items specific for epipolar 
geometry: 
  
create and employ image pyramids for both images in 
order to take the advantage of concept of hierarchy and 
make the algorithms more effective, 
extract sufficient number of features in both images on the 
highest level, 
match features to find candidates for conjugate pairs, 
verify that conjugate pairs of point features are tied 
properly, using relations among them, 
define features more exactly at next pyramid levels using 
cross-correlation, 
estimate orientation parameters, 
  
Figure 10. Total mosaic, 1584 frames, 81 rows 
International / 
- proceed v 
restriction 
- ' estimate 
approach 
The approach 
in digital pho 
Sibiryakov, A 
are shown in 
deviation of t 
respectively. 
  
Figure 11. 
  
Figure 12. 
ori 
Now it’s up to 
meets the requ 
In the preser 
matching for 
point features 
concept of “in: 
are tied accor 
filtered using t 
invariant to im 
found can be 1 
specific work. 
In this study tl 
mosaic creatk 
automatic rela 
In all cases th 
computational