3x3 photos in an analytical stereoinstrument. Various numbers of new 
point observations are added at different stages of the triangulation. 
Single observations are removed in connection with blunder detection. 
The programs were run on a VAX 11/750 at the Departement of Civil 
Engineering at the Norwegian Institute of Technology, Trondheim. CPU 
time was measured for different phases of the computations, such as 
normals formation and factorization, update of normals, and parameter 
solution. 
Results 
Since this test concerns update algorithms, only the CPU-times for 
updating the factorized normals are reported here. 
The diagram below summarises the main trends of the results. It shows 
how the update time increases with the number of unknown parameters in 
the normal equations. Times for TFU and GT are shown for two different 
amounts of additional observations (+2 and +12). Just to get an idea of 
the magnitude in relation to other parts of the computation, time for 
back substitution is shown as well. Note that one "observation" here 
means one observed photo point, implying two observation equations (for 
x and y). 
  
  
  
CPU. 1 
time 
(sec) ——ÓrT. I TFU 
+12 
1.504 UU" ATTUATTOI GT A 
) 
— 1 - UBack subst. A. 
1.00: dA 
o 
"lg 
. 504 
— 
pee TT fa! 
.00 ALL ——— ir rt p , 
T 
50 100 150 200 
# of Parameters 
CPU time versus number of parameters 
= 363