matic and analytical aerial triangulation worth men- 
tioning. In the wake of these differences are some con- 
sequences which need to be discussed further among 
users and developers. 
concept of a point: a point is an abstract quantity 
which does not exist in reality. A tie point or 
control point, for example, is the result of a so- 
phisticated analysis by a human operator, em- 
ploying image understanding and reasoning abil- 
ities far exceeding those available on machines. 
Automatic methods cannot compete and the res- 
cue is extracting features, ranging from interest 
points to edges and to regions. A “point” then 
represents a feature, for example, an edge can be 
represented by characteristic points, a region by 
its centroid. It follows that it would be better 
to deal with the extracted feature as an entity, 
rather than emulating it by points. So the quest 
is to extend block adjustment methods to include 
features as entities. 
number of points: for economical reasons tradi- 
tional aerial triangulation works with as few 
points as possible. Such considerations do not 
apply for automatic methods. The disadvantage 
that automatically determined “points” are less 
carefully selected is compensated for by increas- 
ing the number. It turns out that hundreds of tie 
points per image with a lower accuracy compared 
to manually measured points still render better 
exterior orientation, both in accuracy and relia- 
bility. The motto is: “from quality to quantity.” 
features vs. points: in general, features are more 
tangible quantities than points. Apart from in- 
creasing the robustness of aerial triangulation, 
using features is also desirable in subsequent pro- 
cesses, such as DEM generation or map compi- 
lation. For DEM generation it is advantgeous to 
begin with as much information about the sur- 
face as possible. It may be possible to include 
useful features (e.g., breaklines) in the aerial tri- 
angulation process to aid DEM programs. 
A final note on the differences of the various auto- 
matic aerial triangulation systems that have been de- 
scribed in proceedings and journals. They all must 
solve the essential tasks on which this paper elabo- 
rated intensively. 
initial assumptions: about the exterior orientation 
and the surface of the project area. These as- 
sumptions range from expecting accurate exte- 
rior orientation elements (e.g., from GPS/INS) 
to less demanding assumptions (e.g., "aerial 
case"). The initial assumptions include the cam- 
era model. Apart from central projection, au- 
tomatic aerial triangulation is also successfully 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
   
applied to other applications (see, e.g., Ebner et 
al., 1994). Presumably more severe are the as- 
sumptions about the surface topography: they 
may range from flat to mountainous (e.g., 1/3 of 
the flying height). It should be noted that these 
assumptions are sometimes not explicitly labeled 
as such, rather they are a consequence of the se- 
lected methods. 
selection of tie points: solutions span a wide 
range, from random to planned selection (e.g., 
analyzing location, topography, image content, 
etc.). 
transfer of tie points: some authors combine the 
two processes of transferring and measuring tie 
points though these are two separate processes. 
The transfer entails predicting conjugate image 
locations. As simple as it may appear, it is an 
intricate process that involves error propagation 
from an image into object space and back to 
the target images. Moreover, it should include 
a check whether the feature indeed appears on 
the target image. 
matching tie points: solutions range from pair- 
wise matching to true multiple image matching. 
We conclude this paper by a quote from (Ackermann, 
1995) *...every effort should be made to develop au- 
tomatic digital aerial triangulation. It will be of great 
practical and economic benefit to photogrammetry 
and—in combination with GPS camera positioning-— 
will revolutionize the orientation problem." 
7 ACKNOWLEDGEMENTS 
It is with great appreciation that I thank The Ohio 
State University for granting me a sabbatical year. I 
am grateful to Heinrich Ebner, TU Munich, and Di- 
eter Fritsch, TU Stuttgart, who made it possible for 
me to spend part of the sabbatical leave at their in- 
stitutes. The exchange of experience and stimulating 
discussions with their research teams helped shaping 
up ideas, revising some concepts and confirming oth- 
ers. 
REFERENCES 
Ackermann, F., 1995. Automation of Digital Aerial 
Triangulation. Proceedings 2nd Course in Digital 
Photogrammetry, Bonn. 
Ackermann, F. and V. Tsingas, 1994. Automatic Dig- 
ital Aerial Triangulation. Proc. ASPRS/ACSM An- 
nual Convention, Vol. 1, pp. 1-12, Reno. 
Agouris, P., 1992. Multiple Image Multipoint Match- 
ing for Automatic Aerotriangulation. PhD Disserta- 
tion, Deptm. of Geodetic Science, OSU, Columbus, 
Ohio. 
744 
  
  
  
    
   
  
   
   
   
   
  
  
   
   
   
  
    
    
   
   
   
  
   
   
  
   
   
   
   
   
   
   
   
   
  
   
   
   
    
   
   
    
   
   
  
   
   
   
    
   
  
  
  
  
   
  
   
  
   
    
Bec 
Ein: 
243 
Chc 
Ori 
Scie 
Ebn 
latic 
02/1 
moc 
Ebn 
198’ 
trek 
met. 
Fôrs 
Trai 
PP. 
Frit: 
angi 
PP. 
Frits 
der 
gula 
Gük 
men 
185- 
Hau 
tal / 
'95, 
Heip 
nung 
und 
try. 
Kru] 
Obje 
tion, 
Ohic 
Krzy 
A Ne 
latio: 
215- 
Mayi 
latio; 
225— 
Sche: 
zur F 
traur 
Schei 
tion. 
Schei 
surfa 
Tang