International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
2.2 Ground truth data
Ground truth data were collected in Espoonlahti area. Reference
points were measured with a robotic tachymeter Trimble 5602
DR200+. Altogether 4500 asphalt and gravel reference points
were used in comparison. They were situated in the area of
about 210 by 140 m°. A building map with vectors was
obtained from the City of Espoo. 58 corner points of the roofs
were measured with Leica SR530 Real-Time-Kinematic RTK
GPS. The distance between the rover and the reference station
was about 1300 m during the measurements.
2.3 Comparison of laser points with reference points
Elevations of airborne laser scanner derived points were
compared with the robotic tachymeter reference points. À circle
with a radius of 2 m using a reference point as a centre point of
the circle was created for every reference point. Statistics of the
ALS points were calculated inside the circles if there were more
than 5 laser points included. Mean value, median, minimum,
maximum and standard deviation, nearest laser point to the
reference point and an interpolated height value from the laser
points were calculated. A 10 cm by 10 cm grid and a cubic
method was used in the height interpolation calculations. A
maximum value of 0.2 m for standard deviation inside the circle
was used as a homogeneity measure of the points. Laser
observations from the bushes, trees and buildings are kept out
of the calculations this way. The same method has been used in
(Ahokas et al., 2003).
2.4 Building extraction
Building vectorization was done by using the TerraScan
software developed by Terrasolid Ltd. (www.terrasolid.fi).
Laser points were first classified to ground class. Then building
class was classified by height from ground, e.g. minimum
height for a house must be 2 m above ground. The places of
possible buildings were then known and the building
construction could be started. By pointing a cursor to the
possible building location and clicking the mouse, the program
started to find planar surfaces inside the search area resulting in
a list of planes with the information about boundary colour,
slope angle, number of points matching plane, average
mismatch from point to raw and adjusted plane. User could also
search for additional planes if there were small, undetected
parts in the house. When the correct number of planes was
reached the boundary tools could be applied. The boundary
type of a plane could be set to a rectangle with four 90 degree
corners, to rectangular with 4 to N 90 degree corners or to a free
shape polygon. Boundary shapes could be modified so that a
vertex or a segment can be moved. A vertex could be also
removed from a boundary. Pieces could be cut off from a
corner or a segment. Boundary lines were auto aligned with
intersection lines and with each other. When we were satisfied
with the building the model was applied and the work with the
next building started.
3. RESULTS
3.1 Analysis of height errors
The calculated laser point heights (c.f. section 2.3) were
compared to the heights obtained with the robotic tachymeter.
Firstly, mean height error between the reference point and the
nearest laser point was calculated. Secondly, also the mean
height error between the laser point and the interpolated height
value was obtained. Results are in Tables 2 and 3. Each flight
strip was analyzed separately.
Flight strip 2First | 3First | 4First | SFirst | 6First
Nearest pointz | -1+4 0+4 0+4 E 1+4
Interpolated z -1z:3 0:3 0:3 1:3 1:3
Table 2. Mean height errors (Laser-Tachymeter) and standard
deviations of differences for asphalt and gravel
ground. First pulse mode observations. Results are
in cm. 2First is the strip number 2, first pulse data.
Flight strip 2Last | 3Last | 4Last | SLast | 6Last
Nearest pointz | -244 0x9 0x4 0x4 0x4
Interpolated z -243 0:3 0:3 0x3 1:3
Table 3. Mean height errors (Laser-Tachymeter) and standard
deviations of differences for asphalt and gravel
ground. Last pulse mode observations. Results are
in cm. 2Last is the strip number 2, last pulse data.
It can be concluded that in the reference point area the
systematic error was negligible. Additionally, 3-4 cm standard
deviation is quite small for such plane targets. The decimetre
accuracy obtained with the strip 3 with nearest point method at
last pulse mode is also acceptable.
3.2 Analysis of planimetric errors using centres of buildings
and ridges
An area covering 39 buildings was extracted from the five
overlapping flight strips using methods depicted in section 2.4.
Due to the turnings of the airplane, at least 30 buildings could
be identified from all flight strips. Houses were grouped by
their roof types into five categories: hipped roof, double slope
pitch roof, flat or pitch roof, ridge roof and connected ridge
roof. The extraction of the hipped and connected ridge roof type
buildings required more manual work than e.g. ridge roof
buildings.
Figure 1. A rotated oblique view of the flight strip number 3
(first pulse) laser points and the extracted buildings.
Internati
The use
The dist
and the «
calculate
building;
building;
points of
the build
Flight st
| Min dist
Table 4.
Flight sti
Max dist
Mean +S
Table 3.
Flight si
Min dis
Max dis
| Means
Table 6.
Mean di
were les
when cc
deviatioi
+14-18 «
less than
the acro
boundar
by trees
when us
points €
coincide
Mean ei
from on
extractio