Matching primiti- Method for primiti- 
Area of appli- 
  
    
  
  
  
   
  
   
  
    
  
  
  
  
   
   
  
  
   
  
  
  
   
  
  
  
  
  
  
  
  
  
   
   
   
  
   
Deriche et al. operator of Harris, 
References d Matching method : Availability Comments 
ves ve extraction cation 
; i ; ; different definition 
grey value win- cross correlation aerial and | inhouse development of thomas and mo 
Hannah (1989) dows around Hannah Operator | (left to right and | close range | of SRI International, del E : 
el coordinate sy- 
points right to left) imagery USA y 
stems 
zero crossings of | line matching fol- experimental system 
Schenk et al. ; | B s ; ; ; : 
(1991) lines and points |LoG, Forstner ope-| lowed by least [aerial imagery| at The Ohio State rotation invariant 
rator squares matching University 
Miiller, Hahn : 
feature based mat- experimental system ; 
(1992); Haala et mt Fórst , hi hecked B al i €. Stutteart Umversi model surface as grid 
nts Orstner operator | ching, checke aerial imagery| at Stuttgart Universi- : 
al. (1993); Hahn, per P s e | y Ey e DTM also available 
, cross corrrelation ty 
Kiefner (1994) 
; feature based mat- implemented in 
Tang, Heipke ; T "pcr 
(1993; 1996) points Moravec operator | ching, checked by |aerial imagery| PHODIS ST from autonomous system 
: cross corrrelation Zeiss 
no interior orienta- 
grey value win- cross correlation tion necessary, 8- 
close range | experimental system 
  
    
   
  
   
  
  
  
   
  
   
  
   
  
  
  
  
  
straight lines 
dows around (left to right and ; point-algorithm, Le- 
(1994) s Stephens (1988) 3 imagery at INRIA, France T 
points right to left) ast-Median-Squares 
for blunder detection 
relational des- el and s devel : lof exi 
aerial an research develop- most gener. xi- 
Wang (1994; criptions be-  |different image pro-| relational mat- P ) : o M et 
; : ; 3 close range | ment, Hannover Uni- |sting systems, 8-point- 
1995; 1996) tween points, li- cessing tools ching ; ; ; 
imagery versity algorithm 
nes and areas 
relational des- 
inti H Fórst t tinal ; research develop- 
criptions be- Orstner operator, | relational mat- a ne : et 4 
Cho (1995; 1996) P ; p ; a aerial imagery | ment, The Ohio State | not rotation invariant 
tween points and Burns-lines ching 
University 
  
  
  
  
  
    
    
     
    
    
  
   
   
   
    
   
   
   
     
    
   
    
   
    
Table 3: Approaches for automatic relative orientation 
  
Figure 6: Conjugate points, automatic relative orien- 
tation, example Manhattan, New York City 
(Images courtesy of The Ohio State University) 
4.2 Automatic absolute orientation 
4.2.] How to tackle the problem 
The absolute orientation relates image or model coordi- 
nates to the object space coordinate system. As in the 
case of interior orientation, one of the involved coordi- 
nate systems is only given implicitly. Here, it is the object 
coordinate system, and it is defined through control in- 
306 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
formation which must be provided externally. The task at 
hand is thus semantic. 
In analytical photogrammetry the model surface is avai- 
lable to the operator after relative orientation via stereo 
viewing. An automatic module, however, obviously does 
not have stereo viewing capabilities, and thus, unless the 
surface has been explicitly extracted, e.g. in the form of a 
DTM, it is not available. Therefore, automatic absolute 
orientation has to deal with the relationship between the 
image and the object coordinate system, the model sy- 
stem does not come into play. In case only one image is 
available this task amounts to matching the object space 
control information with features or structures extracted 
from the image. If two or more images are available, the 
extracted image features and/or structures must be mat- 
ched between the images and also with the object space 
control information. These two steps need to be integra- 
ted into one approach in order to yield consistent results. 
Traditionally signalised points were used as control in- 
formation. Giilch (1994b; 1995) has tried to automate this 
approach, but concludes that due to radiometric varia- 
tions of 
the surr 
in pixel 
to imag 
the que 
automa 
In ord 
control 
- geor 
- radk 
- visit 
- well 
=. inde 
-i, inde 
- easy 
= acce 
info 
Contro 
already 
tion in 
- ima, 
fica 
- CON 
- poir 
aine 
Spoi 
= ane2 
== thre 
hou 
The to 
come | 
both c: 
in ord 
proble: 
it mus 
should 
contro 
Torleg 
As me 
autom 
topic « 
howev 
explic: 
orient: