AUTOMATIC BREAKLINE DETECTION USING AN EDGE PRESERVING FILTER
Dietmar Wild, Peter Krzystek
INPHO GmbH Stuttgart, Germany
Mostafa Madani
Intergraph Corporation Huntsville, USA
Commission III, Working Group 3
KEY WORDS: Automatic Breakline Detection, Edge Preserving Filter, Edge Extraction, DEM Generation
ABSTRACT:
Breaklines are important morphological quality features of a digital elevation model (DEM). They represent areas of local maxi-
mum surface curvature and must be adequately described by the geometrical model of the DEM package in use. Most of today' s
software packages for automatic DEM generation utilize breaklines which have been measured in advance by the human operator.
This purely interactive process is time consuming and hence costly. In order to overcome this drawback, we propose a procedure
which automatically detects breaklines during the DEM generation process. The method would reduce the interactive input of the
operator to a minimum and increase the economical efficiency of automatic DEM generation.
Basically, conventional DEM filter methods are disadvantageous if they are applied to sharp height changes of the terrain. They do
not take into consideration structures in the data and therefore they may also lead to a loss of information. Edge preserving filters
are more flexible and largely avoid the elimination of breaklines. The advantage of the implemented method in MATCH-T is the
detection of local surface areas of maximum curvature. In such areas the spacing of the DEM posts is automatically decreased and
hence, discrepancies between the true terrain and the DEM are minimized.
The paper presents a new approach to breakline detection which is based on the principle of adaptive edge preserving filters. It out-
lines the procedure and its application to DEM breaklines. Practical results are presented referring to an open pit coal-mine area
and an area of mountainous terrain. It is shown in particular, how breakline points and breakline directions are automatically de-
tected in breakline areas.
1 INTRODUCTION
Digital elevation models (DEM's) are used to represent the
surface in a reduced amount of data. The main information
about a surface is represented in structure lines (e.g. break-
lines). To minimize the model error caused by the spacing of
the DEM posts and the discrepancies between the DEM and
the true surface it is necessary to record the structure lines as
3D-polylines. This information is used to fit the DEM to the
real terrain and to reduce the spacing of the DEM posts in the
breakline areas.
A number of software packages for automatic DEM generation
can handle breaklines, but all approaches have in common that
the breaklines have to be measured in advance. If they have no
knowledge about the morphological structure, the DEM is glo-
bally smoothed since sharp height changes of the terrain can-
not be detected. In order to overcome this drawback, we
present a method of DEM generation which preserves and de-
tects breaklines using an adaptive edge preserving filter in a ro-
bust finite element adjustment.
The investigated approach uses curvatures and torsion as addi-
tional observations in the least squares adjustment. The local
maximum surface curvature and the local breakline direction
are derived from the residuals of that curvature and torsion ob-
servations. The terrain discontinuities are preserved by weight-
ing the curvature and torsion observations in dependence on
their residuals and considering the breakline direction during
the subsequent filtering process. The weighting of the observa-
tions is based on two weighting functions and is able to elim-
inate gross errors in the data. Hence the influence of the gross
errors is minimized.
946
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
The results of the adaptive edge preserving filter are DEM
posts that lie approximately at breaklines. The accuracy of the
approximation is largely dependent on the spacing of the DEM
posts. The first goal of the adaptive edge preserving filter is the
detection of local surface areas of maximum curvature. At
such areas the spacing of the DEM posts is automatically
adapted and hence, discrepancies between the true terrain and
the DEM are minimized.
The next development step is to subsequently tie up the DEM
posts with the maximum terrain surface curvature as 3D poly-
lines by using the breakline direction that is also assessed by
the filter. The connections of the approximate breakline points
to structure lines represent the first geometrical approximation
of the surface structure. Those initial breaklines derived solely
in object space are verified against gray value edges in image
space which are derived by applying a gradient operator to the
images. The final breaklines are evaluated by fitting an object
model (e.g. spline) to the breakline points corresponding to
gray value edges.
The approach presented here avoids to initially start the break-
line detection from image space since gray value edges does
not necessarily correspond to breaklines. Instead, breaklines
are analyzed in object space using an edge preserving filter
method and are tracked back into the images by verifying
them with gray value edges.
The paper outlines the mathematical background of the adapt-
ive edge preserving filter for the DEM generation. The DEM
and the addition breakline information are the basis in the con-
cept of the automatic breakline detection. The practical results
of the examples represent the state of development of the auto-
matic breakline detection using an edge preserving. filter
method.
2.1
ha:
val
for
fin
an
19
ide
ist
sin
x
CO;
sel
ac
tio
en
wi
to
the
Sp:
ar
