Peter Lohmann
Figure 4: Results of Filtering (Left column: Original data; Mid: Linear prediction; Right: Dual Rank)
The result of morphological filtering shows a quite reliable recognition and removal of buildings and artificial
structures, which are marked by a clear height-difference to their surroundings. These objects are separated very
precisely on the basis of the implemented algorithm. Surfaces, which ascend or descend in nature diagonally and show a
continuous slope, like dams or dikes, remain unconsidered in the first step of the algorithm, using the standard deviation
filter. Consequently, these types of terrain features remain unchanged in principle. Break lines are first marked as edges,
but do not form closed features. The resulting very narrow, long shape is excluded from further processing by analyzing
the size and the form of a ROI, as a function of the biggest existing building. In case of filtering the vegetation one has
to distinguish between single objects and extensive vegetation. Individual trees or bushes, exceeding the predefined
height-difference to their surrounding, are recognized unequivocally and removed. Problems emerge with low
vegetation which cannot be separated from ground points by examining their heights alone. In areas of dense
vegetation, particularly in forests, the procedure of the laser-scanning leads to a wide scattering of height-values caused
by the reflection at different vegetation-horizons. Therefore, ground-points and points of low vegetation are difficult to
separate. For this reason, forest areas are removed nearly completely, leaving only few ground-points in comparison to
the method of linear prediction.
The interpolation of the removed regions through bilinear interpolating sufficiently adapts to the surrounding terrain.
The typical network pattern caused by this interpolation is minimized by averaging to an extent, that it is only
perceptible at larger filled up regions. This is the case at big buildings like factories and also at extended woodlands.
Due to the fact, that this method leaves most of the topography unchanged (selective filtering), the accuracy is the same
than that of the original measurements in non-filtered areas.
546 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
Eck
pp 4
Frit
and
Vor
Auf
Vel
Hui
scar
261
Pho
Kili
Pho
Koc
Inst
deu
Loh
Pro:
and
Loh
IVA
Loh
Ver
MV
Sch
Pho
deu!