409
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
Figure 6. Sorted ratios of the variation coefficients after vs.
before normalization of the intensity.
About 75% of the selected regions become values below 1.0
meaning the new coefficient is better (smaller) than the
previous one. In some cases with higher values than 1.0 the
regions contain chimney and dormer windows. On the other
side we can not be sure, that the borders of the regions are well
defined inside a homogeneous area.
Obviously the effect of intensity normalization is not as much
evident for street regions. The variance of the incidence angle is
much smaller than for roofs.
5.4 Intensity of a region with different geometry
For the investigation on the intensity within a region, we select
two neighboured planes with the same material and the same
gradient direction but varying roof slopes. The intensity values
for all points inside this region are visualized in
Figure 8 coloured by the flight number.
Figure 8a shows the original data and the approximating cosine
curve as black line. In b the normalized intensity values are
scaled in such a way, that the mean value, drawn by a black line,
remains the same as before. The correspondence between flight
number and colour is depicted in top of
Figure 8b. There exist no points from the flights 1 and 6.
5.3 Visualization of the Normalized Data
The intensity improvements are demonstrated by the following
figures showing the intensity values before and after the
normalization by the incidence angle. For comparison reasons
the colours dark blue and dark red are bounded to the thresholds
5% respectively 95% as lower and upper percentiles of the
intensity. The normalized intensity reflects higher intensities
without large variations for the roof planes but lower values for
points near the ridge, where the planarity is not given. The
results for a pyramidal roof including four planes are shown by
Figure 4. The original data shows higher values for the south
west planes than for the north-east ones caused by the flight
paths and directions (
Figure 2). In the normalized data all four planes have same
intensity values and appear homogeneous. A building
composed by several parts with different orientation is given in
Figure 7. The original data demonstrates again the dependency
of the intensity from the incidence angle. By the normalization
this dependency is compensated and again the intensity is
adapted.
Figure 7.Intensity data for different orientated roofs.
Gabled roofs: a) original, b) normalized, Pyramidal roof: c)
original, d) normalized.
10 20 30 40 50
incidence angle [°]
t • • » * • « •
c 5000 . • ...
•— 4000
"O
<D
3000—-
CO
§ 2000
o
1000,
• i: ‘A . •« X vt;
■■ ü
L AfT i.. J
10 20 30 40 50
incidence angle [°]
Figure 8. Intensity values vs. incidence angle coloured by the
flight number: a) original data, b) normalized data.
From all other flights we observe always two cluster caused by
the roof of the main building and an extension of it with
different normal vector. Both belong to the same region. To
give an example, for flight 7 we have incidence angles for 25°
and 50° and for flight 5 for 6° and 31°, as examples. The total
number of points inside this region is 1055. The number of
points belonging to the different flights can be seen in
Table 2. The ratio of the variation coefficients for this region is
R v = 0.68 , subsequently the normalized intensity value is 32%
better than the original one.
Flight
2
3
4
5
7
Points
77
156
259
308
255
Table 2. Number of points in the selected region measured by
the corresponding flights.
Based on the high variation of the intensity before and after
normalization by the cosine law, we suppose, that the influence
of surface effects like the kind of material or local geometry can
not be ignored for man-made surfaces. In contrast to this Lutz et
al. (2003) observe high variations for the intensity values for
natural materials.
5.5 Adaptation of the Lambertian Law
The reflectance may not be always in accordance with the
Lambertian law. Adapting the Lambertian cosine law and trying
to minimize the variation coefficient for each region separately
by varying the power p of cos(<9) p the best values are
achieved for powers between 0.5 and 1.1 (Figure 9).