Thomas Vögtle
ising The big advantage of this vegetation index is, that it is hardly affected by seasonally changing appearance of the plants
es of in the data sets. A binarization of the NDVI values leads to an image containing only vegetation areas (Figure 3).
se of
In a last step mathematical morphology (Serra, 1982) is used to remove obvious inhomogenities caused by shadow
effects and spectral deviations. In this case, a closing and an opening is applied (one dilation, two erosions and finally a
n be dilation) to eliminate small areas consisting of only one pixel and to obtain more homogenous vegetation areas (Figure
data 4). However, the unfavorable smoothing effects of segment boundaries may cause artifacts in the subsequent vegetation
;) by reduction process.
DVI
jects
(DVI
> and
13D
lanes
o the
ne or
Figure 3: Segmented vegetation areas by means of NDVI Figure 4: Unified vegetation areas by means of
morphological operators
rence 3.2 Laser scanning data
ically
For detection of 3D object on the surface of the Earth, terrain undulations have to be eliminated. Therefore, a special
filter has been used for extraction of terrain points (DTM) out of the laser DEM (von Hansen & Vógtle, 1999). After
calculating a difference DEM (dDEM) out of laser elevations and this extracted DTM only 3D objects on the surface
(e.g. buildings, trees, bushes, cars etc.) remain and all terrain points have a value of about zero. To eliminate the still
present vegetation objects now a superimposition of this dDEM (Figure 2) and the vegetation image (Figure 4, sect. 3.1)
is carried out and every dDEM element which belongs to a vegetation area is set to ground level (= zero). A detailed
comparison of the status before and after this reduction is shown in Fig. Figure 5. Obviously a certain number of
artifacts can be observed. These are caused by residual errors in geometrical rectification, smoothing of the boundaries
of extracted vegetation areas by morphological operators and shadow regions inside vegetation areas. However, these
disturbances can be eliminated in further processing steps using shape, size and plane characteristics.
hilling
Figure 5: Reduction of vegetation areas; left side: test building 1, right side: test building 2
(1)
For separation of individual objects as building hypotheses a region growing algorithm was implemented. This merges
neighbouring pixels of dDEM which have only small elevation differences and lie significantly above ground level. The
algorithm stops at high gradients which occur normally at the boundary of each 3D object, assigns an individual
segment number and therefore determines the membership of each dDEM pixel to a certain object (Figure 6).
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 929