The test site nr 1 (Stuttgart)
The highly urbanized area with a steep slopes, mixture of
vegetation and buildings on hillside. The distance between the
laser measured points ranges from 1 to 1.5m.
ME 150 Re
lf 120000
5403650
s r
MI 5419500
5410850
194000 494050 494100 424100
Figure 4. The test site 1 and 2
The test site nr 2 (Vaihingen)
This test field is located in a rural terrain, of steep slopes and
lush vegetation. The distance between the laser measured points
ranges from 2 to 3.5m.
4. RESULTS
4.1 The accuracy assessment
The set of a reference data
The reference data was generated by the hand-controlled
filtering of the cloud of laser points. In the filtration process the
knowledge about the tested area, and aerial photographs were
used. After the hand-classification all the classified points were
properly assigned to two groups: bare earth, and object.
The quantitative assessment of data
Considering the possibility of the easy comparison of presented
filtration algorithm with the algorithms used in the ISPRS tests
(see paragraph 4.2), it was decided to use the same method of
the accuracy assessment.
In the accuracy analysis there were considered two types of
errors. The Type I classification error appeared, when the
reflecting laser point was assumed to be a terrain feature point
instead to be properly classified as a ground point. In opposite,
the Type II classification error was noticed when a laser-
scanned point was wrongly assigned to the ground points,
instead to the feature points.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
Filtered
Bare Earth Object
- Bare Earth 14195 6880
Q 4
&l3[ obj 558 [4386
o
c L9] Bare Earth 11973 626
L
oic
o Object 198 3494
Table 5. Results of filtering for site 1 and 2. Bold — number of
Type II errors. Underlining — number of Type I
errors.
For the groups of errors of Type I and Type Il, there were
calculated the following statistical functions of deviations of the
estimated values from the true values: the mean, RMSE, and
magnitude of the maximum and minimum error (Tab. 6).
% Min. Max. Mean | Std Dev.
[m] [m] [m] [m]
Sie l| Type] | 326] -46 5.9 -0.1 9:7
Typell | 37 | -1155 17.8 -1.3 2.8
3.7
Site 2| Type! | 50 | -L28 1.33 -0.01 0.36
Typell | 5.4 | -5.14 1.68 -0.63 0.92
Table 6. Results of filtering for site 1 and 2.
4.2 Comparison with the selected filtration algorithms
The directives of the WG III/3 "3D reconstruction from
airborne laser scanner and InSAR data" gave in 2002 a push to
the research project concerning comparison of the existing
methods of automatic laser data filtering (Sithole, Vosselman,
2003). The main objective of that research project it was
determination of functionality of the filtering algorithms in a
certain field conditions (mainly the configuration of
topographical surface, and different types of the terrain
features).
The presented algorithm was compared with the eight methods
reported to the ISPRS test. To get the reliable comparison
factors, the same accuracy measures were used. The graphical
visualization of the comparison results is shown on the Fig. 7
and Fig. 8.
70 mr
60 |
30 | B m m " [M |
20 " | d
10 í, 4 p |
0 | à s — 2 |
ES BD FE hn religums
= O =
E85 do aeo But
£ + 5 0 => 5 +
x 0 He
LL X a LL.
B Type errors À Type ll errors
Figure 7. Comparison of the reliability factors for the Site 1.
The errors Type-I and Type-II are shown in % for
each of the considered filtration algorithms
Inter]
4.3
last
In o
on t
anal
scan
impi
field
impi
dist:
Uni
bet
ofr
imp
pro
the