ul 2004
--->
1g (for-
egment
ard, ct.
eeds in
1st sur-
its. In
in the
ab) Ran
on the
ion).
v patch
atil the
nt (e.g.
surface
f one
ne ini-
d. For
imated
fferent
vature
of the
sented
e ALS
tion of
n anal-
nation
of the
sakline
cedure
sented
| addi-
ration
rocess
he re-
by the
alt für
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
° E2=-0.0002
Figure 7: Determination of the breakline direction
(eigenvector of the smallest eigenvalue, in this example
E2) with the help an adjusting quadric on the basis of the
ALS point cloud.
5.1 Modelling based on a 2D Breakline Approxi-
mation
The example in this section presents the result of a 3D
breakline modelling procedure based on original unclas-
sified ALS data. Figure 8 shows the intermediate steps
within the modelling procedure starting with a rough ap-
proximation of the breakline (upper part of the image). In
the next step the boundary for the data selection and the
ALS point cloud within this area (black dots) can be seen.
In the lowest part of the figure the refined 3D modelled
breakline is presented. In this example the size of the sur-
face patches was 5m (along the breakline) by 10m (across
the breakline). The overlap between neighbouring patches
was 50 percent.
Figure 8: Work flow (beginning at the top) of the 3D
modelling based on a 2D approximation of the breakline
demonstrated on a practical example. The middle part
shows the buffer zone around the breakline approximation
and the ALS point cloud (black dots) inside this area. In
the lowest picture the refined robustly estimated breakline
is presented.
1101
5.2 Modelling based on 3D Breakline Growing
The demonstration of the 3D breakline growing (cf. sec-
tion 4) is presented on the previous breakline example.
The growing in this example (cf. figure 9) is performed
into both directions. The final resulting breakline can be
inspected in the lowest part of the figure. The break off
point is defined by an intersection angle smaller than 10
degrees. The patch size is 10m by 10m with 50 percent
overlap between neighbouring patches. This results in a
breakline description with a point distance of 5m. The
patch size for the growing is bigger than in the previous
example in order to make the process more robust. The
result can be refined (if necessary) in a further modelling
procedure.
Figure 9: Breakline growing (beginning at the top) based
on a manual digitised start segment.
5.3 Integration of Breaklines within the Terrain
Determination Process (Filtering)
A small example for the consideration of breaklines within
the filtering process is presented in figure 10. The derived
DTM shows a high quality enhancement in the areas of
sharp ridges due to the explicit 3D modelling of the break-
lines. For the DTM generation robust interpolation (cf.
(Kraus and Pfeifer 1998)) was used. Within this process
the previous determined breaklines were considered to be
free of gross errors. The final DTM of this example has a
hybrid raster data structure.
6 SUMMARY
This paper presents a method for 3D modelling of break-
lines based on the original unclassified ALS point cloud.
The modelling is performed with the help of robustly esti-
mated surface patch pairs along the breakline. The elim-
ination of the influence of off-terrain points within this
estimation process works in a fully automated way and
adapts itself to the data. This modelling starts from a 2D
approximation of the breakline, which is iteratively refined.
Additional observations (e.g. acquired using image data)
can be easily integrated into the modelling procedure. A
further section focuses on methods for the automatisation
of the whole process.
