mments 
ıt definition 
nage and mo- 
ordinate sy- 
stems 
n invariant 
urface as grid 
Iso available 
nous system 
rior orienta- 
ecessary, 8- 
gorithm, Le- 
lian-Squares 
ider detection 
:neral of exi- 
tems, 8-point- 
zorithm 
tion invariant 
  
. The task at 
face is avai- 
n via stereo 
viously does 
s, unless the 
he form of a 
atic absolute 
between the 
e model sy- 
ne image is 
object space 
es extracted 
vailable, the 
iust be mat- 
object space 
) be integra- 
stent results. 
3 control in- 
iutomate this 
netric varia- 
tions of the signal, poor contrast between the signal and 
the surrounding image content, and the small signal size 
in pixel units, 'new concepts for ground control, adapted 
to image analysis’ are needed. This conclusion leads to 
the question, which requirements control information for 
automatic absolute orientation has to fulfil. 
In order to be reliably extractable from the imagery 
control information should ideally be 
- geometrically well defined, 
- radiometrically unique, 
- visible from various directions, 
- well distributed across the imagery, 
- independent of image content, 
- independent of image scale, 
- easy to represent in two and in three dimensions, 
- accessible (the object space coordinates of the control 
information need to be determined). 
Control information which can be (and to some extend 
already has been) used for automatic absolute orienta- 
tion includes 
- image chips showing e.g. road crossings or other signi- 
ficant structure, 
- complete orthoimages, 
- point-like objects such as manhole covers, 
- linear objects such as roads, rivers, and the corre- 
sponding networks, 
- area-type objects such as land parcels, lakes or forests, 
- three-dimensional wire frame models of objects, e.g. 
houses. 
The topographic objects used as control information can 
come from a GIS or a scanned and vectorized map. In 
both cases it must be ensured that the data are up to date 
in order to separate the orientation from the revision 
problem. If the third dimension is not explicitly available, 
it must be provided in form of a DTM. In this context it 
should be noted that the DTM itself can also be used as 
control information (Ebner, Strunz 1988; Rosenholm, 
Torlegárd 1988; Ebner, Ohlhof 1994). 
As mentioned, there exists a close connection between 
automatic absolute orientation and the more general 
topic of image analysis. Automatic absolute orientation, 
however, is easier, because the control information is 
explicitly given. A generic strategy for automatic absolute 
orientation consists of the following steps: 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
  
- select appropriate control information, 
- define the primitives to be extracted, taking into ac- 
count the appearance of the control information in the 
images, 
- design an algorithm for extraction of the primitives, 
- match the primitives with the control information, 
- compute the orientation parameters according to the 
predefined model. 
Thus, the task remains very complex. It is argued here 
that a general solution for automatic absolute orientation 
will not be available in the foreseeable future. Therefore, 
approaches tailored to more specific applications should 
be investigated in which additional assumptions can be 
introduced: 
- The image content restricts the selection of possible 
control information. Images of urban scenes contain 
many man-made objects consisting of straight lines, 
and thus the extraction and matching algorithms can 
be based on these features and relations between 
them, e.g. distances and angles. However, many occlu- 
sions and shadows have to be expected. 
- The image scale dictates the representation and to 
some extend also the selection of the control informa- 
tion. In large scale aerial imagery point-like objects 
might be depicted large enough to be reliably detec- 
ted. As another example, roads should be detected on 
the basis of edges. As the image scale decreases more 
generalised representations of the same control infor- 
mation must be used, e.g. centre lines for roads. Image 
scale also influences the type of transformation. For 
example, if satellite imagery is to be geo-referenced, a 
two-dimensional solution might suffice to fulfil the 
given accuracy requirements. 
- Prior knowledge of the orientation parameters is usu- 
ally available from a flight plan or direct measure- 
ments (see discussion on GPS/INS above). This infor- 
mation should be incorporated in any solution to auto- 
matic absolute orientation. 
4.2.2 Examples for automatic absolute orientation 
In this section solutions for automatic absolute orienta- 
tion suggested in the literature are presented and discus- 
sed. The presentation is ordered according to the list of 
possible control information (see above). The intension 
is not to give a complete but a representative list. 
307