ve 
lata 
the 
the 
bic 
the 
> of 
on), 
idi- 
the 
ind, 
nes 
and 
] of 
rea 
nce 
S to 
1 of 
ing 
1 of 
ola- 
the 
rect 
for 
ists 
lan- 
that 
tion 
le - 
10le 
nish 
ing 
yoth 
area 
SN. 
e of 
sted 
the 
not 
ping 
The adjustment of the SN is done by the program system 
ORIENT (Kager, 1989), due to its universal possibilities in 
adjustment matters. The required data (a planar, possibly 
cyclic SN-graph and the LSP-sequences) have to be extracted 
(with the help of a ‘rover’ moved along constraint-edges) from 
the triangulation for each SN and have to be converted into a 
structure suitable for ORIENT. Before a SN can be extracted, 
all those candidates of junctions of constraint-lines have to be 
detected which are not contained obviously in the captured data 
and are therefore not triangulated as LNKs until yet (due to 
gaps stemming from the data capture process). 
  
Fig. 7: Lines ending with a junction at another line or 
crossing another line. Because the junctions 
have not been captured directly with a LSP, 
they have to be searched for during the ex- 
traction of the actual SN. 
Subsequent to each adjustment of a SN with ORIENT, the 
quality of the adjustment is verified. As criterion of quality the 
deviations of the LSPs from the corresponding adjusted curves 
are used. If the deviation from the adjusted curve exeeds three 
times the root mean square error of the observations at at least 
one LSP, a further adjustment of the SN with a condensed 
arrangement of SKs follows. (Until now it is assumed, that 
blunder-detection of the LSPs has been done before the SNs 
are adjusted.) This loop of optimisation is repeated until the 
demanded quality is reached, or a further condensation of SKs 
becomes impossible. 
After the adjustment of a SN, the courses of the original con- 
straint-edges have to be removed from the triangulation and, 
correspondingly, the adjusted ones have to be triangulated 
anew. In addition, the user shall have the possibility to judge 
and edit the results of the adjustment in a graphical way, 
before a new triangulation of the adjusted SN follows. 
5 IMPLEMENTATION AND RESULTS 
The surface is modelled - as described above - by decomposi- 
tion into simple objects and the determination of the relations 
among those. This object-orientated concept imposes an object- 
orientated implementation. The main attributes of these objects 
are the adjacency and incidence relations among them. 
Thereupon bases an important concept of the implementation - 
the rover - concept: Rovers contain references to few data- 
objects, and perform, under use of the objects relations, local 
operations on these objects. E.g. a 'triangle-rover' contains 
references to the three vertices of a triangle and performs 
operations, such as : 
' changing to the neighbouring triangle 
calculating the triangle-normal 
inserting of a point into the triangle 
positioning on the triangle, nearest to a given point 
A rover only works locally and always processes few and ad- 
jacent data-objects. 
The presented concepts were implemented and tested. Espe- 
cially the various optimization-criterions were examined in 
regard to their characteristics and properties. 
Results of the presented methods are shown in figure 8 and 9. 
  
  
  
  
  
  
  
Fig. 8: Triangulation of a sand-pit. About 360 points have been measured. A combination of minimizing the maximum angle 
and maximizing the angle between two adjacent triangles has been used for optimization. 
411 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996