tanbul 2004
), Northing
tics on the
heights and
istogram is
itive height
' vegetation
> compared
| be filtered
ie 17 check
,
3. O0 [m]
+ 5.74
n 5235 tie
h tie points
ysition and
r the look
hus, it was
parameters,
s only.
only
o data sets.
lering pure
; data none
is not that
PC for the
| (less than
adjustment
ose results
ed 7 check
a only, all
o [m]
6.05
à. 5233 tit
adjustment
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B1. Istanbul 2004
From the analysis of other three-line-scanner imagery as e.g.
the German MOMS-02 camera we already know, that the
presence of a nadir view primarily improves the horizontal
accuracy and does not directly affect the height accuracy
(Ebner et al., 1992). This also can be observed comparing the
results of the tables 5 and 6 with the tables 3 and 4. The
statistics on the height differences remains more or less the
same, while the horizontal differences at the 17 check points
increase - but still remain far better than 1 pixel.
3.5 DSM by strict model
After the adjustment the imaging geometry is known and
image points can rigorously be transferred into object space
using the estimated internal and external orientation
parameters. From the resulting 3D mass point cloud TIN
models are produced, which later are exported into 10m
raster DSMs.
3.5.1 Mass point cloud transformation into object space
The mass point cloud is transformed point-wise into object
space by a local adjustment based on equation [6], which is
the inverse form of equation [1] using the estimated model
parameters listed in table 2. For each point and image 3
equations are defined:
2 (U) x
X=X,+tu R,'R
Y=Y +4 Ry Ry -(u)y [6]
Z-Z tu: Ry: Ry);
Using the X, Y, Z object coordinates and the scale factor pt as
unknowns, 9 equations are formulated to solve for 6
unknowns (X, Y,Z.1, 15,15) in the case of 3-ray-points, and 6
equations for 5 unknowns (X,Y,Z,t,15) in the case of 2-ray-
points. The transformation is d dd separately for the
above mentioned 3 point groups: 2-ray-points, 3-ray-points
matched in 2 and in 3 combinations. The test sites #1, #3, #7
and #8, which contain control points are excluded from the
following accuracy assessment.
3.5.2 Comparison with the reference DTM
After the transformation the height coordinates are compared
to the reference DTM heights, which previously have been
interpolated for the respective horizontal coordinates. In
figure 5 the height differences of the mass points for TS #5
are depicted in a color coded representation:
blue: dh « -5m
— lightblue: | -5m«dh «-3m
= green: -3m < dh < 3m
— Orange: 3m < dh < 5m
= red: dh > 5m
As can be seen, the matching algorithm works quite well, if
the image contains sufficient contrast and texture. On the
other hand, in those parts with poor contrast and/or texture
the matching failes, causing gaps in the 3D point cloud and,
consequently, in the DSM. In the orthoimage in figure 6 the
pointless areas can clearly be identified as areas with low
contrast (like the wide road in the central lower image part)
or homogeneous texture (like the forest in the upper right
corner of the image). It also can be seen, that red points,
indicating height differences bigger than 5m, mainly appear
in the urban area in the left part of the image or in forest
zones, e.g. in the upper right image part. This illustrates well
the difference between the surface model deduced from the
point cloud and the reference terrain model representing the
bare Earth’s surface. Table 7 shows statistics of the complete
comparison between point heights and reference DTM.
Figure 5: Color coded height differences wn points and L
reference DTM for TS £5 (section: 5.3 x 3.6 km)
min 6: rd te of TS #5 AES. 5, 3 Te 3. L6 km) |
TS N Min Mean Max RMS c
42 601042 -27.5 0.3 30.2 3.8 3.8
44 . 485450 -85.5 0.3 88.0 3.2 5.1
#5 678165 -49.4 1.5 33-9 4.0 T7
#6 446380 -39.7 109 66.9 13.0 72
Point group 1: 3-ray-points from 3 matching runs
42 164411 -69.1 0.1 56.3 4.1 4.0
#4 119301 -1422 1.6 1654 11.6 11.4
#5 86635 -1090 1.1 99.0 5.1 5.0
#6 68979 -106.7 8.1 95.9 12.4 9.4
Point group 2: 3-ray-points from 2 matching runs
#2 134013 .-175.3 0.1 99.8 5.4 5.4
#4 57965 -141.1 1.7 2036 13.8 15.7
#5 103744 -223.6 1.3 226.3 6.0 5.9
#6 35571 1516. 66 151.9 124 10.6
Point group 3: 2-ray-points
Table 7: Statistics on height differences dh [m] between the
3D object points and the reference DTM
The results are derived from the complete unfiltered sets of
the automatically matched points, including blunders and
points on top of vegetation and buildings. For time reasons
no filtering or editing was done. In this respect, the RMS
values must be interpreted as conservative with a
considerable potential for improvement. Since TS #6 covers
the Barcelona urban area and contains only few ground
points, the accuracy potential of SPOTS should rather be
deduced from the results of the other test sites #2, #4 and #5.
