The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
roof planes and hence the intensity values are higher for the
eastern planes and lower for the west ones.
3.2 Point Density
The calculation of the incidence angle and the planarity is based
on the determination of the covariance for each point by
including all neighbour points inside a sphere with predefined
radius. For a radius of lm we may get about 30.5 points as
average, if we include all flights. Flight 3 delivers as average
only 8.8 points. Increasing the radius by factor 2 we get the
average value of 120.6 points per sphere and all flights included,
where 33.6 points are originated from flight 3. Due to the small
size of the roof regions we select a radius of lm to avoid too
much disturbance for the points near the border of the object
planes. The inhomogeneous and anisotropic point distribution
causes a reduction of the planarity value and raises its standard
deviation even if the 3rd eigenvalue equals to zero, because the
eigenvalues belonging to both eigenvectors parallel to the plane
may differ, as discussed by Gross (2006). Investigations
concerning the quality of the point position are also presented
by Bae et al. (2005).
4. SELECTION OF HOMOGENEOUS REGIONS
For the assessment of the adapted intensity / we prefer regions
with different orientations but homogenous surface reflectance
to separate the influences of the incidence angle and material
effects. The roof planes within our scene cover a large band of
possible incidence angles but most of them have same tiles. The
selected regions contain the same material but varying angle vs.
flight direction and the direction of the laser beam. Each roof
plane is labeled by a region number.
region number region number
a b
Figure 5. Angles [°] of flight 3 for all selected plane regions
sorted by the mean angle together with its standard deviation: a)
slope of roofs, b) incidence angles.
This selection includes a wide range concerning the off nadir
angle for the laser beam. The variation inside the regions is
small because the regions are small in comparison with the
distance to the sensor. The slope angle of the roof planes (
Figure 5a) encloses a few nearly flat roofs but also steeper roofs
up to 50°. For each point of the point cloud inside the region the
slope angle is calculated based on the eigenvector of the
smallest eigenvalue. Therefore the data set encloses regions
with small and height variations of the slope angle, which may
be influenced by small objects on top of the roofs.The planarity
yields high values for planes objects, where the mean value
varies from 0.67 to 0.83. Due to noise and disturbing small
object parts, higher values could not be achieved. The standard
deviation inside the regions varies from 0,06 to 0.13, which
indicates, that the planes are not exactly planar and does not
show the same roughness. The incidence angle (
Figure 5b) varies from 2° to 68° with a mean value from 44°.
The standard deviation delivers values from 0.5° to 12° with a
mean value from 4°. Inside a region the variation of the
incidence angle is small. The distances R between sensor and
object surface varies from 429m to 449m with a mean standard
deviation of lm.
5. RESULTS
For the selected regions the given intensity is normalized by
division with the cosine of the incidence angle. By this division
the normalized intensity value increases compared to the
original one. Therefore we use the mean value /u(x) and the
standard deviation cr(x) for the calculation of the variation
coefficient V c (x) = cr(x)/p(x) . This coefficient is scale
invariant and regards the dependency of the standard deviation
from the intensity as presented by Pfeifer et al. (2007).
5.1 Global Consideration Over all Regions in Common
Mean value and standard deviation of the variation coefficient
over all roof regions with nearly the same material
p(V c ( region ) ) and a ( V c ( region ) )
are determined and written in
Table 1. Considering only flight 3 or 4 there are no significant
value modifications, but including flight 3 and 4 together the
normalization delivers an essentially smaller standard deviation.
The variance of the incidence angle for each region increases, if
data from more than one flight are used. In the last column of
Table 1 we see the corresponding values by regarding all flights.
In this case we get a good improvement for the normalized
intensity.
Flights
3
4
3-4
1-7
before
normali
zation
MKÏ
0.145
0.142
0.162
0.189
a ( V c) ■
0.030
0.027
0.028
0.039
after
normali
zation
M(r c )
0.145
0.144
0.150
0.161
A v c)
0.027
0.023
0.023
0.023
Table 1. Mean value and standard deviation for different flight
situation data sets before and after normalization.
5.2 Consideration of each region separately
For an assessment the ratio of the variation coefficient
R v (region) = V c after (region) ¡V cbefore (region) for all
selected regions after vs. regions before normalization are
calculated. The sorted ratios are drawn in Figure 6.
408