ul 2004
aber of
Type I
C were
| of the
E, and
] Dev.
[m]
0.7
2.8
).36
).92
wa
from
ush to
isting
Iman,
| was
s in a
n of
errain
thods
arison
phical
Fig, 7
ès i
i
o
=
=
m
1.
x
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
60
50 l-4———— ———— ——————————— ese:
40
30 i 4
10 mA |o -.,
A | n i A E
0 Fe y, uw À
D C Le = Oo D —
eue ea 5 RE m ux
gv # $ o 2 = =;
u = X LC
m Typelerrors 4 Type llerrors
Figure 8. Comparison of the reliability factors for the site 2.
The errors Type-1 and Type-II are shown in % for
each of the considered filtration algorithms
4.3 The algorithm modification — intensity of the first and
last reflection
In our algorithm the filtration errors appeared in the first place
on the forest fields.Therefore/a special effort was made to
analyze some additional data collected during the laser
scanning. The intensities of the first and the last returned
impulses were compared for the area of the selected small test
fields of the site 2. Also the spatial distance between those two
impulses was calculated. It was noticed that for over 45% that
distance was smaller than 0.5 m. (see Fig. 9).
percent
0 5 10 15 20 25 30
distance
Figure 9. Histogram of first-last impulse distances for forest
area.
Unfortunately, there was not discovered any precise correlation
between the distance of the first-last impulse and the category
of reflecting surface (bare earth or object). Only for the first-last
impulse distance exceeding 15 m (Fig. 10)., we can with high
probability assume, that the last impulse was really reflected by
the bare earth.
N
we
LL] bare earth. |]
Bl object
percent
distance
Figure 10. The relationship between the firs-last impulse
distance, and the type of surface (bare earth, object) for forest
area
In the next step of our research, there was analyzed the
relationship between the intensity of impulse reflection and the
type of reflecting surface. It was expected that the surface type
(object — trees or bare earth) can be recognized by the intensity
of reflected impulse. In the investigated case the intensity value
of returned pulses ranged from zero to 190 relative units (see
Fig. 11). Only for the intensity value greater than 190 relative
units (Fig. 11) the points reflecting the last impulse can be
qualified, with a high probability, as a bare earth.
[1 bare earth
BH object
T T T
percent
ü 1 m 1 0 1 fl L L
0 50 100 150 200 20 300 350 400 450 500
intensity
Figure 11. The relationship between the last impulse intensity
for the bare-earth and object reflecting surfaces for
forest area.
5. CONCLUSION
In this paper a FFT based method of filtering of airborne laser
scanner data has been presented. The method is directly derived
from the signal processing theory. It is so, because we introduce
the laser points location as an independent variable, and we
treat the terrain and the terrain features represented in a discrete
form (DEM), as a discrete signal that records the elevation
variations for data points in a x- y location.
The results from presented algorithm were compared with both
reference data and with filtering results of eight methods
reported to ISPRS test. Our experimental results reveal that
quality of derived DTM is quite high. This algorithm allows for
separation of the urbanized areas with high reliability. However
filtering of forest areas has been the most difficult problem.