IS: 
jr matching, 
eous adjust- 
of the addi- 
e differences 
combination 
uris (1992)]. 
; among the 
patches are 
he first and 
ind the third 
second and 
formation to 
iy values as 
'ences. The 
) as the gray 
luced as un- 
ch matching 
sequence of 
re than two 
images (or 
described in 
and its use 
known, it is 
s area to ob- 
bed here the 
of the recti- 
are not real 
ces between 
cantly smal- 
rationale of 
here for the 
1e extension 
o images is 
metric situ- 
ie conjugate 
pr is an ap- 
as not lie on 
e surface el- 
lori, but are 
procedure), 
iarped. The 
warped im- 
: d, again, on 
proximation 
s the differ- 
aces become 
ound by the 
ATEGY 
terative pro- 
(S: 
age patches, 
ording to the 
    
   
  
  
  
   
   
shift (found by matching) 
Figure 1: Schematic description of the concept of matching 
warped images. 
approximate approximate DEM approx. 
orientations tie points around tie points 
  
    
  
  
last 
iteration? 
Figure 2: Outline of the proposed matching scheme. 
improved tie points 
  
improved orientations 
   
   
true surface 
corrected location 
approximate surface 
  
additional matched 
points 
  
  
improved DEM 
around tie points 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
  
warped 
patch n-1 
(tp = tie point) 
  
   
  
  
   
  
     
       
  
     
    
  
  
   
  
   
   
new tp, 
add. points 
photo 0 
new tp, 
add. points 
photo n-1 
Figure 3: Warping and Matching Phase. 
available approximations of the orientation parameters 
and the surface around the point. When no surface ap- 
proximation is available, a horizontal plane is assumed. 
During the final iteration, a multiple-patch matching 
is performed with these warped image patches to de- 
termine the exact image coordinates of the tie point. 
In all iterations but the last one, a grid is formed in the 
object space, centered on the approximate tie point. 
Multiple-patch matching is used (in the same manner 
as in the last iteration) to match each point on this grid. 
Each matched grid point is projected back, through 
the available surface, to the image space. As explained 
earlier, the resulted photo coordinates are more accur- 
ate than those from the previous iteration. The ob- 
tained photo-coordinates are used for the "Block Ad- 
justment Phase." 
Block Adjustment Phase (Figure 4): The . photo- 
coordinates of the center of the grid found in 
the "Warping.and Matching Phase" constitute the 
input for a bundle block adjustment. The results of 
the block adjustment are new (improved) orientation 
parameters. 
DEM Update Phase (Figure 5): Using the new orienta- 
tion parameters estimated in the "Block Adjustment 
Phase," a surface is reconstructed around each tie 
point intersecting the photo-points (from the "Warping 
and Matching Phase") back to the object space and in- 
terpolating them into regular grids. The grid interval is 
half the interval that was used in the former iteration. 
Since each “Warping and Matching” phase yields better loc- 
ations for the conjugate photo-points, the results of the block 
adjustment render improved orientation parameters. These 
orientation parameters, together with the improved conjugate 
points, lead to better approximations for the object surfaces 
around each tie point. The process converges iteratively to 
the desired solution.