MULTI-TEMPLET-MATCHING FOR THE AUTOMATION OF PHOTOGRAMMETRIC MEASUREMENTS 
Prof. Dr O. Kölbl, U. Bach, D. Gasior, K. de Laporte 
Swiss Federal Institute of Technology, Lausanne - Switzerland 
ISPRS Commission III 
SUMMARY 
The algorithm of Multi-Templet-Matching is a matching technique for photogrammetric measurements. It repre- 
sents the basis for the derivation of digital terrain models but also for the automation of aerotriangulation or the 
elaboration of orthophotos. An analytical plotter is used for the scanning and for the verification of the matching 
results. The paper presents the matching algorithms and the computer environment. 
KEY WORDS 
Automatic height measurement, matching, digital terrain model, scanning of aerial photographs, digital photo- 
grammetry 
1. INTRODUCTION 
Image correlation and the appropriate digitizing of aer- 
ial photographs play a key role for the automation of 
the photogrammetric working process. If one succeeds 
in automating the matching of corresponding sections 
of two aerial photographs, then one can develop pro- 
cedures allowing a high degree of automation of the 
photogrammetric restitution. This concerns primarily 
the derivation of digital terrain models, but such a 
procedure can also be used as a starting point for the 
automatic derivation of digital orthophotos, which 
should play a key role for all planning activities in the 
near future. Furthermore, digital orthophotos are in- 
creasingly used in geo-information systems. If one un- 
derlays the vector information on the monitor of a 
GIS with an orthophoto, then most of the photo- 
grammetric plotting operations can be done directly on 
the working station. Otherwise, image matching can 
also be used for the automation of aerial triangulation. 
For the last few years, a number of research groups 
have concentrated their efforts on image matching. 
However, one gets the impression that only partial re- 
sults have been achieved. Even if practical solutions 
are currently offered in close-range photogrammetry 
and for the correlation of small-scale photographs, 
only partial solutions are available for large-scale pho- 
tographs. At the Institute of Photogrammetry of the 
EPF-Lausanne, a small research group has been heav- 
ily engaged for some years in the problem of image 
matching and close collaboration was achieved with 
the firm Leica, Unterentfelden. This commitment led 
to the development of a reliable procedure for image 
matching, which can also serve as a basis for com- 
pletely automated aerotriangulation. 
One of the key elements for the automation of image 
restitution is the algorithm for image matching. 
Moreover, the technique of image digitizing and the 
instrumental equipment are of the utmost impor- 
tance. At the very beginning, the Institute experiment- 
ed with the dynamic programming, which served as a 
basis for the development of the Multi-Templet- 
Matching. In a first phase of this algorithm, the paral- 
laxes of square image segments of varying size are de- 
termined. These parallaxes of overlapping picture 
540 
elements are then introduced in an adjustment proce- 
dure by which the terrain is approximated by finite 
elements. The procedure also allows the introduction 
of break-lines and the incorporation of discontinuities. 
In this way, the terrain is approximated, including ir- 
relevant objects like houses or trees, which are subse- 
quently largely eliminated by filtering. Finally, the cor- 
relation results are verified and edited before integra- 
tion into a GIS station. 
Beside the height parallaxes, the correlation procedure 
also gives the Y-parallaxes of the picture elements. 
Consequently, the procedure also supplies the parame- 
ters for the computation of a relative orientation. If 
these elements are taken together for a whole block 
formation, one can even automate aerotriangulation. 
These procedures are implemented on the instru- 
ments of the Institute of Photogrammetry, where the 
digitalization is done on an analytical plotter DSR15. A 
parallel processor based on Transputer technology has 
been developed for the computer-intensive correla- 
tion algorithm; this processing is done independently 
of the digitalization on the analytical plotter. Subse- 
quently, the results of the image correlation are trans- 
ferred to the analytical plotter for verification. In this 
context, the analytical plotter still plays a key role for 
data processing. However, while it is clear that the 
analytical plotter could be replaced by a digital plotter 
for the verification, it is yet not evident that the verifi- 
cation process can be done more efficiently on a digital 
plotter than on the analytical plotter. 
2. DIGITIZING OF 
AERIAL PHOTOGRAPHS 
The quality of the digitalization process is of the ut- 
most importance for automatic image processing. It is 
essential that the whole range of grey levels of an aer- 
ial photograph be converted into a computer-readable 
form with minimal noise. If not, it is quite impossible 
to expect from image processing about the same results 
as from visual photogrammetric measurements. 
In a first approach, it does not seem very easy to estab- 
lish tolerances for the digitizing of aerial photographs. 
Most logically, one will start to analyse the efficiency of 
the human eye as visual procedures are well proven 
tio 
tre 
va 
ize 
res 
na 
Fig 
Sen 
ceri