International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
presents a new method for three-dimensional modelling of
breaklines. It is based on a pair wise intersection of over-
lapping robust surface elements (determined with the help
of irregular distributed points) along the breaklines. Af-
ter an introduction into the basic concept the paper fo-
cuses on the adaption to the modelling of unclassified ALS
data, which allows the determination of breaklines even in
wooded areas. Furthermore, the integration of additional
data (e.g. information from digital images) into the mod-
elling process is treated. For the modelling a 2D approx-
imation of the breakline is essential. Therefore a further
section concentrates on procedures for the automatic de-
termination of these initial values. An examples section
demonstrates the practical use of the presented methods
and the integration of breaklines into the classification (fil-
tering) process for DTM determination. À short summary
with an outlook on further development issues concludes
the paper.
2 STATUS OF RESEARCH
Up to now the research in the arca of breakline extrac-
tion from ALS data concentrated on the development of
methods for the fully automatic 2D detection of breaklines.
For this aim image processing techniques were usually used
(c.g. first derivatives (Gomoes-Percira and Wicherson 1999),
gradient images (Sui 2002), Laplacian operator (Gomes-
Pereira and Janssen 1999)). A summary of some of these
and other methods and an additional new method based on
hypothesis testing (homogeneity measure: quadratic vari-
ation) can be found in (Brügelmann 2000). In general
these raster based algorithms are applied to a previously
generated (filtered) DTM. The result of these detection
methods are pixels marked as edge pixels. In a further
raster to vector conversion 2D breaklines can be gener-
ated. 'This conversion includes at large some smoothing
in order to eliminate zigzag effects caused by the raster
structure. The height of the breakline is independently
extracted from a slightly smoothed vegetation free DTM
at the planimetric position of the detected breakline.
Published experiments show quite interesting results for
the 2D extraction. Especially vegetation free breaklines
on dikes can be extracted quite well in 2D, but certain
breaklines are only partly detected or some of them were
not detected at all. Another problem of these methods
is that they operate only in 2D and just interpolate the
height from a more or less smoothed D'TM, which can be
eventually affected by classification errors caused by the
DTM generation process (see remarks in the section 1).
The height of the breakline is computed totaly independent
from the determination of the 2D position. However, for
a high quality modelling a method which allows a full 3D
determination of the breakline within one process has to be
preferred. A basic concept, which allows a 3D refinement
of approximately 2D detected or just manually measured
breaklines, was already published in (Kraus and Pfeifer
2001) and (Briese et al. 2002). In the following sections
extensions based on this basic idea, allowing the modelling
of breaklines on the basis of unclassified original ALS point
clouds, are presented.
1098
3 BREAKLINE MODELLING
mm
I'he modelling should be performed on the basis of the
original point cloud in order to use all possible information
and to exclude preprocessing steps, which can lead to a
quality reduction (e.g. re-sampling of the point cloud into
a regular grid or classification (filter) errors during DTM
determination). Additionally to a high quality modelling,
the algorithm should deliver accuracy measures in order to
get an estimation about the quality of the 3D modelling.
In the following subsection the basic principles of the break-
line modelling method are presented. Step by step the
method will be extended in order to cope with specific
ALS problems (off-terrain points are considered in subsec-
tion 3.2). Furthermore the integration of additional in-
formation sources (e.g. image data) is considered in sub-
section 3.3. The basic modelling concept presented in this
section is a semi-automated breakline modelling procedure
for which a rough 2D approximation of the breakline is nec-
essary. Concepts, which allow to overcome this limitation,
will be presented in section 4.
3.1 Basic Concept
Basically, the breakline is described as the intersection line
of continuously overlapping analytic surface patch pairs
(compare figure 2). The determination of the patch pairs
is performed with the help of a simultaneous adjustment
of both analytic surface pairs supported by the point cloud
data within a buffer zone around the approximative break-
line. The access to the ALS points within this buffer zone
is performed with the help of a topographic database. Be-
fore the intersection the point data is classified on the basis
of the rough 2D approximation of the breakline into left
and right point groups.
Left patches
Figure 2: Basic concept for the description of breaklines
with the help of intersecting patch pairs determined on
the basis of surrounding point cloud data; Left: Ground
view of overlapping patch pairs; Right: Perspective view
of a reduced number of patch pairs (overlapping pairs are
removed) with their point cloud support.
For the left and right patches we used intersecting plane
pairs, which should approximate the left and right sur-
face tangent plane as good as possible. Within the 3D
modelling procedure the approximation of the breakline
is refined step by step (this leads to a reclassification of
Interr
the j
of th
recog
a inc
funct
Curve
breal
far a
certa
to th
get t
(we
point
funct
aroui
fluen
usual
to a
influ
of th
P >
weight [0:1]
*
for e
line;
point
Blue
of th
buffe
weigh
As a
point
terse
patch
sectic
break
that |
justm
unkn
lows !
aceur
propa
3.2
As m
with |
cloud
we ha
ence (
