fol. XXXVIII, Part 7B
In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
95
i using the kNN algo-
osest point pair enters
distance dij between
; smaller then the min-
e time the 3D distance
cm), (4)
umber of laser points,
cal points are consid-
xpected that the mean
iroximately zero.
t density, compared to
easons. First, from an
be arbitrary close to-
while the vehicle can
ually more laser scan-
e at the same time. It
it drive-lines can have
sition time is different
y have changed. Also
fferent quality. There-
points are investigated
e data set.
rent scanners (scanner
verlapping drive-lines
;ntical points are ana-
possibly identify sys-
i geometric attributes,
points is investigated.
erpolate a DTM. The
:rse Distance Weight-
>han and Toth, 2008).
topics, however they
The main emphasis is
i a number of individ-
Huaxing, 2008). The
terrain points (FD1),
t (FD2), terrain point
FR) and interpolation
ted from the existing
)rs (FD1, FD2, FD3)
’he fourth influencing
:d to the interpolation
zero (Kraus et al., 2006). First a grid of 1 x 1 m size is laid over
the terrain laser points. For grid cells, which include 4 or more
terrain laser points, a tilted plane is modeled in a Least Square
sense by a first order polynomial as given in Eq. 5:
Z = do d - a\X + a 2 Y. (5)
Here X, Y, Z are the coordinates of the terrain laser points (ob
servations) that are included into the plane computation and 00,
ai and a2 are the unknown plane coefficients. The graphical rep
resentation of each term in Eq. 5 is shown in Fig. 3. To make the
Z Y/
■z 'it
/ (X.Y.Z Æ
/ '7/
/fM,zyy
/ V
z=a 0 x
Z=ajX X
Z=a,Y
Figure 3: A graphic representation of terms given in Eq. 5; after
(Li et al., 2005)
least squares computation more efficient, a new coordinate sys
tem is used with the interpolation grid point (Xg,Yg) as the ori
gin; therefore the method is called Moving Least Squares (MLS)
adjustment (Karel and Kraus, 2006). The equation of a plane
(Eq. 5) simplifies, so the plane coefficient ao becomes the eleva
tion of the grid point itself, as written in Eq. 6:
Zg = ao (6)
The mathematical model of Moving Least Squares for linear sur
face fitting is then given in matrix vector notation as in Eq. 7b:
y TU A- X (7a)
' Z\
’1 Ax -X G Yi - Yg
z 2
1 X 2 - Y 2 - Y g
ao
=
a i
(7b)
a 2
Zn_
1 X n — Xg Y n — Yg_
Where Xi, Y i: Zi for i = 1... n are the coordinates of the n
original laser terrain points included in the plane computation.
Then the unknowns in vector x and their variance-covariance ma
trix Sài are computed in a least squares adjustment as written in
Eq. 8 and Eq. 9 respectively:
¡6 = (A r E- 1 A)“ 1 A T E- 1 y (8)
E** = (A T E^A) 1 (9)
Where Y yy is the variance matrix of observations. Here, the the
oretical height precision of the laser points azi computed in Sec
tion 2.2 is used. Besides, the vertical distances between the orig
inal terrain points and the modeled plain, are computed. Those
residuals e are applied to calculate the Root Mean Square Error
(RMSE) as written in Eq. 10 for each plane:
RMSE =
(10)
precision of the original laser points (FD2), their density (FD1)
and distribution (FD3). The second term a\ represents the qual
ity loss due to the representation of the terrain surface. In this
research the RMSE is considered as a measure of the terrain sur
face roughness (FR) with respect to the plane modeled by the cho
sen random-to-grid MLS interpolation (FI). Therefore a e simply
equals the RMSE as computed in Eq. 10.
3 RESULTS AND DISCUSSION
In this section the results of the quality evaluation of both LMMS
laser point heights and a derived DTM are discussed for a LMMS
data set representing a stretch of Dutch coast. Before these results
are given, first this data set is described in more detail.
3.1 Data description
The LMMS data set was acquired on the Dutch coast near Egmond
aan Zee using the StreetMapper system owned by provider Geo-
maat (StreetMapper, 2010, Geomaat, 2010). The acquisition took
place on November 27, 2008 at the time of low tide. Within 2
hours a stretch of beach of 6 km long and 180 m wide was cov
ered. The point cloud consists of about 56 million laser points.
As experienced by Geomaat, the 3D laser point coordinates and
the classification into terrain and non-terrain points can be done
within 2 days. In this research a smaller representative test area
of 213x101 m was chosen, which is covered by 8 drive-lines,
see Fig. 4. The data set consists of 1 220 825 laser points. Each
record of a laser point has 15 attributes, which are: 3D laser point
position X, Y, Z, intensity I, class number C, scanning angle
0, time of point acquisition T, drive-line number DL and scan
ner number SC, range R, incidence angle a, footprint diameter
Df p , range error due to scanning geometry 5R, measuring pre
cision az,m and geometrical precision az,sr- The second data
set used in this research composes of eight trajectories positions
within the test area (black lines in Fig. 4).
x 10 s
5.1651
5165
5.1649
^ 5.1648
E
ra 5.1647
c
'■§ 5.1646
Z 5 1645
5.1644
5.1643
5.1642
1.0316 1.0318 1.032 1.0322 1.0324 1 0326 1.0328 1.033 1.0332
Easting [m] x 10 5
Figure 4: The digital photo of the test area [GoogleMaps]. The
black dashed lines mark the trajectories driven downward i.e.
from the north to the south and the solid lines mark the trajec
tories driven in the opposite direction.
3.2 Results of height differences of identical points
)06), a grid point ele-
lear interpolation (FI),
es and co-variances of
to estimate the quality
ly speaking the preci-
a standard deviation
ors are assumed to be
To finally predict the DTM quality uotm, a mathematical model
after (Li et al., 2005) as written in Eq. 11 is used.
o 2 dtm=°1 0 +o-l (11)
Here the standard deviation of the constant plane coefficient o ao
represents the quality of the original data and accounts for the
By analyzing the attributes of the identical point pairs, i.e. the
scanner and drive-line number, it is concluded that the major
ity of identical point pairs belongs to the same scanner and the
same drive-line. Fewer identical point are found in the scanner
overlap and drive-line overlap (see Table 1). In Table 1 the re
sults of height differences for the three cases are presented. The
mean (avg) of height differences AZ is very close to zero for the