International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
2.1.1 Pseudo-orthophotos 
The rectification of a single aerial photograph yields an 
orthophoto, which can only be called “true” if the DTM used is 
correctly representing the terrain surface. On the other hand, a 
rough DTM, which is only approximating the surface, can be 
used for the rectification of each stereo partner. This gives a 
new set of images, which we call "pseudo-orthophotos" (POPs). 
These images contain small deviations 4R from the "true" 
orthophoto caused by the erroneous heights of the rough DTM 
(see Fig. 1). However, they have the same overall scale and 
orientation, which is an important property for facilitating 
matching. In combination with the rough DTM, POPs still 
contain the same stereo information as the original photographs, 
which enables strict photogrammetric 3D reconstruction. Using 
POPs for automated digital point transfer (matching) instead of 
the original photo scans has several advantages: 
e Perspective distortions are removed to a great extent. 
e — Area-based matching methods, e.g., cross-correlation 
and least-squares matching, show an increased 
accuracy and robustness because of the more 
favorable geometry of POPs. 
e Only the area of interest, i.e., the rock glacier and its 
surroundings, is covered by the POPs, therefore less 
computer storage is needed. 
e 3D perception of motion parallaxes in multi-temporal 
POPs is possible, giving a first impression of the rock 
glacier creep and other surface changes. 
  
Rectification 
& 
Reconstruction 
  
  
  
  
  
  
  
  
  
  
  
POP, . Matching 
AR, | RP 
POP, : 
P, 
DTM  ... true digital terrain model 
(DTM) ... rough digital terrain model 
AR; ... radial displacement in POP; 
Figure 1. Object point reconstruction with POPs 
As mentioned before. a pair of POPs can be used for strict 
photogrammetric point reconstruction in object space. For the 
corresponding points P, and P; in the POPs, the height is 
interpolated in the rough DTM, which yields the intersection 
points D, and D», respectively. The projection rays passing 
through the projection center C, and C» coincide with the rays 
used in the rectification process and also allow for forward 
intersection of the object point Q. This concept has already been 
proposed for DTM generation (Schenk et al., 1990) and more 
generally outlined for the extraction of 3D information from 
rectified stereo pairs (Baltsavias, 1996). 
It is important to note that deviations of the rough DTM from 
the true terrain surface do not introduce any systematic errors 
for the reconstructed object point position (however, increasing 
height deviations will cause additional perspective distortions 
resulting in reduced matching accuracy). If, on the other hand, a 
single orthophoto is used for the extraction of a 3D point 
position, the radial displacements of points not lying on the 
given DTM are not corrected for. Thus, POPs derived from a 
DTM of good quality should be used for digital point transfer if 
high photogrammetric precision is needed. 
2.1.2 Constrained matching using the MPCM algorithm 
Multi-photo (geometrically) constrained matching (MPCM) was 
introduced by Griin and Baltsavias (1988). This algorithm 
combines least-squares matching (LSM) in multiple images 
with simultaneous determination of the associated object point 
in a single, iterative adjustment process. The unknown object 
coordinates and the matched image positions found by LSM are 
connected via the collinearity equations. Fixing a selected 
(interest) point in the reference image forces the corresponding 
points in the other images to lie on their epipolar lines. This 
epipolar constraint reduces the search space from 2D to ID, 
thus making the MPCM algorithm more robust and also more 
accurate than the free, unconstrained LSM. A detailed 
description and extensive investigation of the algorithm can be 
found in Baltsavias (1991). 
Because of its robustness and high matching accuracy the 
MPCM algorithm is considered to be a perfect tool for 
photogrammetric rock glacier monitoring. It is implemented as a 
single module in the current version 2.0 of the ADVM software, 
replacing both the (unconstrained) LSM module and 
reconstruction module of the prior version. The standard 
MPCM algorithm had to be adapted (1) for multi-temporal 
matching and (2) for matching in POPs. 
2.4.5  Multi-temporal point transfer in image sequences 
In a rock glacier monitoring project stereo photographs of at 
least two epochs must be available. A prominent point (e.g. a 
boulder) on the rock glacier surface can be tracked in the image 
sequence as it moves due to rock glacier flow. MPCM can be 
used for this multi-temporal point transfer but must be extended 
for simultaneous determination of the object point in every 
epoch. This can be done by adding three additional object 
coordinate unknowns for each new epoch to the MPCM system. 
For such a multi-temporal MPCM system, e.g., containing two 
epochs, the unknown object points Q; and Q» are part of the 
parameter vector. After the system is solved, the displacement 
894 
Int 
vec 
the 
The 
add 
Col 
vec 
pro 
des 
In | 
per: 
POI 
by. 
arbi 
Fur 
whi 
way 
2.3. 
The 
give 
yiel 
into 
the | 
Inst: 
met] 
roug 
tang 
with 
adve 
Usin 
strai 
wheı 
The 
at pc 
to th 
AY). 
X- ai 
the p