2r 
ere 
TEST OF ALGORITHMS FOR SEQUENTIAL ADJUSTMENT 
Knut Ragnar Holm 
RUNIT - The Computing Centre at the University of Trondheim 
Trondheim, Norway 
ABSTRACT 
The performance of systems for on-line triangulation (OLT) as well as 
for real-time point positioning depends on both effective software and 
effective hardware. Although the developement of hardware, e.g. array 
processors, brings about significant progress in speed, there will 
always be a need for the most effective software, in order to fully 
exploit the potentials of the hardware actually used. 
In this context a search for the most effective algorithm for 
sequential adjustment is underway. Some algorithms are proposed, and 
some estimates have been made on the performance, mainly in terms of 
algebraic operations counts. 
However, in an actual implementation of OLT, the performance of a 
specific algorithm also depends on the software structure as well as 
the data structure used. Although the data structure should meet the 
algorithm's performance needs, the software structure should reflect 
the operational procedures of the system. Consequently the performance 
of such algorithms can best be judged with tests in an OLT-like 
software environment. 
The paper discusses the implementation and a comparative test of two 
specific algorithms, namely the Triangular Factor Update algorithm, and 
Givens Transformations. 
INTRODUCTION 
At the University of Trondheim there has for some years been going on 
research on photogrammetry for industrial applications (/12/,/7/,/8/). 
During the FOMAKON project, described in /6/, the actual application 
was production control on large steel structures. Procedures and 
software have been improved, and the advantages of photogrammetric 
measurements have been recognised. Nevertheless, the method has not yet 
been adopted for regular use at shipyards, who were the main potential 
users. 
The main reason for this reluctance is the turnaround time. With film 
developement, photo measurement, computation with blunder removal, and 
production of final results, turnaround time would normally be some 
days. Practical applications often require turnaround measured in 
hours, or even minutes. 
Some kind of on-line photogrammetric triangulation (OLT) would bring us 
+ 357--