International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
the reference data less accurate, compared to Waldkirch. Still, MMO, with an RMS
AIM delivered better results, especially along discontinuities on
Test dato! umbo: Type of points/number of E RMS Mean Maium Std. dev.
of matched points A System with sign absolute
(SS/AIM} comparison points (m) (m) (m) (m)
W1052A/ BE {Pts}/40 SS 1.29 -0.82 3.26 1.00
{292157/166009} AIM 0.56 -0.22 1.00 0.52
BE {Pts, Brklin}/2500 SS 0.95 -0.75 2.80 0.60
AIM 0.70 -0.49 1.69 0.44
BE (hPts}/43 SS 227 -1.90 4.78 1.26
AIM 22 -1.56 3.47 0.90
MMO {Brklin}/1000 SS 1.20 0.09 4.11 1.18
AIM 0.66 -0.02 2.32 0.66
W1052B/ BE {Pts}/30 SS 0.86 0.64 1.84 0.68
1141512/342692} AIM 0.58 0.34 7 0.50
BE {Pts, Brklin}/1100 SS 0.37 -0.10 1.84 0.33
AIM 0.45 -0.36 1:27 0.30
W1052C/ BE {Pts}/37 SS 0.72 0.43 2235 0.59
{736749/283956} AIM 0.55 0.33 1.09 0.45
BE {Pts, Brklin}/24840 SS 1.38 0.19 15.05 1.36
AIM 0.89 0.24 7.42 0.86
BE (hPts}/334 SS 1.79 -0.26 19.24 1.77
AIM 1.23 -0.46 4.66 1.14
MMO {Brklin}/10000 SS 0.96 0.33 7.89 0.89
AIM 0.60 0.17 2.83 0.58
Y0624C/ BE {Pts}/33 SS 4.55 -1.96 14.31 -1.96
(128950/40750} AIM 2.45 -1.71 4.99 1.78
BE {Pts, Brklin}/6553 SS 4.10 -2.65 19.09 3.10
AIM 3.28 -2.50 9.80 2.13
MMO {Brklin}/41738 SS 7.56 1.43 103.96 7.42
AIM 1.30 0.30 7.87 1.26
Table 4. Error statistics of reference data minus automatically generated DSMs. Single and interpolated points from
breaklines are indicated as Pts and Brklin respectively. Points on bare earth (BE) close to building outlines are
indicated as hPts.
error of 1.30 m compared to 7.56 m of SS. In addition, SS
exhibited gross errors of over 100 m. The good performance of
AIM at MMO is also shown by the smaller mean values
compared to SS. The mean values for both methods show that
matching generally measures higher than the manual
measurements, while for MMO matching results are lower. The
fact that the accuracy was less for BE than MMO for Y0624C
and W1052C is due to the density of the buildings and the
relative short distances of the measured points to neighboring
MMO. BE breaklines in these areas have been extracted along
discontinuities on the ground, which were close to MMO.
Therefore, it is possible that modeling errors were introduced.
Figure 6. Elevation errors of 900 reference points along a
specific breakline on MMO minus automatically
derived points with AIM and SS. Larger
differences are observed (dark line) for SS
406
compared to AIM.
The example in Fig. 6 shows the elevation differences computed
for 900 reference points along a specific breakline defining a
roof edge. The differences are smaller for points derived with
AIM, compared to SS, because the use of contour points and
edges in conjunction with the multi-image matching approach
used in AIM, improves the modeling of discontinuities.
4. CONCLUSIONS
In this study, two systems have been evaluated regarding DSM
generation using ADS40 imagery acquired over two different
test areas. The AIM system showed better performance,
compared to SS system, especially along building
discontinuities. Accuracy increased, often by factor two or
more, and blunders decreased, even in difficult areas, as
Yokohama. Various components of the AIM method, especially
those that take advantage of the particular characteristics of
ADS40, lead to this improved performance. With the exception
of breaklines in W1052C, the performance of AIM in the rural
areas, even at MMO, was close to the theoretically expected and
the accuracy of the manual measurements. However, dense
urban areas still pose a problem and need algorithmic
improvements but also denser matching and use of more images
from neighbouring strips that should have a high overlap. À
better configuration of the lines on the focal plane, having e.g. at
Internai
least tw
forward
benefic
AIM is
adaptivi
on Lev
extract:
This wc
of ETH
Tachiba
Corp. f
Vassilie
referenc
general
the assi
LGGM
acknow
Baltsavi
200
and
Pho
Bacher
Gen
35-4
Fricker
coll
Wet
Fritsch
Pho
Frits
Wic
Gwinne
Neu
Exp
DE!
on
Mar
Haala ^
perf
pho
IAP
Henrics
Attr
No.:
Zuri
Hinsker
200
OR]
156-
Hoffmai
auto
acqi
a
Leberl \
200:
Syst
RO!
Neukun
and