DYNAMIC WINDOW SIZE LEAST SQUARES MATCHING FOR AERIAL 
TRIANGULATION POINTS 
Shue-chia Wang, Sysh-Hong Chiu, Chi-Chang Tsai 
Department of Surveying Engineering 
National Cheng Kung University 
Tainan, Taiwan 
Inter-Commission Working Group lI/III 
KEY WORDS: Triangulation, Matching, Automation, System, Design 
ABSTRACT: 
When only the measuring of the aerial triangulation points is concerned, there is no need to digitize the whole frame of the 
aerial photographs, because the triangulation points locate always at the specific so called Von Gruber positions. It is 
sufficient to digitize only a small area at these positions. But to automatically pin point where should be digitized in the 
overlapping photographs is very difficult. Therefore we developed a semi-automatic solution using the Rollei RS1 digitizer 
to digitize only a small area where the triangulation points are supposed to locate (the standard position). 
Within each of these small areas enough triangulation points should be selected and transferred by image matching 
technique into the corresponding areas in neighboring photographs. The selection of the points is done automatically by 
interest operators. The matching of these points between different photographs is done by the Least Squares Matching 
(LSM) method. In order to obtain more reliable and accurate result a procedure using dynamic window size and checked by 
the cross correlation before and after the LSM is proposed. 
In the dynamic window size matching, instead of one fixed size, several different sizes of window will be used to determine 
which one gives the best result and to check if the selected result is reliable. Whenever no reliable result could be obtained, 
  
the system switches automatically to manual selection of the matching windows. 
1. Introduction 
In the development of the automation in Photogrammetry 
by digital image processing, the automation of the 
measurement of aerial triangulation points (TPs) is a very 
promising one, because the measurement of the TPs does 
not need any sophisticated image understanding. It can be 
done by the low level gray value matching alone. No 
knowledge about the image content i.e. no recognition of 
the objects is needed. Therefore there has been great 
success in the past few years in studying the problem 
associated with it and in developing proper commercial 
systems to conduct the automatic TP measurement in the 
practice. The Digital Comparator Correlation System from 
Helava for example, was the first commercial system for 
fully automatic measurement of the TPs (Helava, 1987). 
The same system is used now in the Leica-Helava digital 
photogrammetric workstation DPW. 
Since low level gray value matching is good enough for TP 
measurement, the most difficult task in the development of 
such fully automated system is not the matching itself but 
the finding of the conjugate positions in different 
overlapping photographs, i.e. to find the overlapping areas 
between the neighboring photographs within and across 
the strips . Without this knowledge the search work for 
transferring TPs into overlapping photographs would be 
intolerably large and the chance of false matching would 
also be too big. In the Leica-Helava DPW system for 
example, the operator has to give the information about the 
approximate layout of the strips, the beginning and ending 
of each strip, etc., so that the system can calculate the 
relative position of all the photographs in the block. But in 
mountainous areas in spite of this information, the system 
fails to find the correct conjugate areas very often because 
of the large relief displacement. In such case human 
operator is always needed. In the MATCH-AT system, 
developed by the INPHO Stuttgart based on the research 
result of Tsingas (Tsingas, 1992), the approximate layout 
of the block should also be known before the system can 
start the matching. In the Image Station from 
INTERGRAPH there is still no automatic measurement of 
the TPs. The points have to be measured manually like in 
the analytical plotters. From all these examples we can see 
that a blind, fully automatic measurement of the TPs 
without any prior knowledge of the block layout or the 
approximate position of each photograph is very difficult. 
Even in the future when every camera position could be 
given by the GPS navigation system, the problem of 
mountainous area still needs to be solved. 
In the Surveying Engineering Department of the National 
Cheng Kung University there is a RS1 image digitizer from 
Rollei, which is designed to digitize each time a small patch 
of the photograph by its CCD camera with 512x512 pixels. 
Since for the measurement of TPs, it is not necessary to 
396 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996 
dig 
aut 
use 
me 
bec 
by 
are 
hur 
the 
bec 
loc: 
phc 
the 
aw 
Afte 
ther 
our 
ado 
gre: 
fixe 
Lea 
relie 
cha 
aree 
DCC 
Witl 
com 
one 
aeri: 
Sinc 
very 
the i 
usec 
Aftei 
up. | 
ther 
high 
ther 
used 
TPs 
in ce 
Thes 
the 
TPs; 
Whil 
have 
appr 
accu 
false 
itself 
wind 
wind 
positi