istics
nding
)oints
were
eight
but it
ale of
rance
% of
vel to
lering
o the
are in
rithm
ize. If
ze is
'e is
his to
jon is
efault
and
vith 7
meter
tively,
nts lie
lation
ds of
of the
the
tistics
in are
rer -
been
imeter
tor is
cribes
. The
s. On
s are
d and
above
orithm
nes of
3000
2500 .|
2000]
1500 |
No. of Points
1000 4
emul [7
erssopemm | | |
sicorEBS |
07 «S1
0'zosoqy fm
B t oL to
SATIN Tn =
V ay my
1 1 1 o 9
m m es : Un
in o tà 2
Height Difference in Metres
Fig. 5 Histogram of Height Differences
Min/Max Template Sizes 7 and 11 - 7 and 9
These are particularly dependent on the terrain in the
imagery, with an increase in Parallax Bound and a
decrease in Epipolar Line Distance required with an
increase in terrain diversity. This effectively increases the
search area of the algorithm.
Comparing results from the three against each other, Flat
and Mountainous show the greatest differences, Hilly
and Mountainous are closely related and Flat and Hilly
lie in between. With the nature of the terrain in the
imagery being quite varied and undulating, it was
expected that the three generations would be markedly
different if compared against a benchmark surface.
Comparisons against the semi-analytical generation show
remarkably similar statistics, with all three obtaining
approximately 6496 of points within + 1m of the semi-
analytical attempt.
6.2.2 Urban
Due to problems with the loading of the urban imagery,
testing has been limited on the ImageStation and no
results had been obtained at the time of submission of
this paper.
7. CONCLUSIONS
Clearly there is a large number of possible variations in
the accuracy of the automatic DEM produced when we
consider there are in the order of 16 and 28 parameters in
the two systems. As has been shown in this paper, even
what appears to be quite a small parameter change can
have a significant effect. It must be appreciated by users
from all backgrounds that the variations in results are
significant and that resultant DEMs may in fact be
significantly incorrect.
Investigation is continuing into the points which have
been raised in this paper, including comparisons with a
benchmark surface, integrity of results, specific
Successful/unsuccessful terrain characteristics for certain
parameters and the effects achieved by altering other
361
parameters. Parallel to these investigations will be
research into the effects of scanning resolution,
compression techniques and resampling of imagery
scanned at differing resolutions.
7000 ME
6000_
5000
$
=
S 4000.
e
eu
QU" 3000
e
z
2000
1000 —
0 4 m
1 1 i 1 e ec po ok
POS 505658 $4 $5
déco de ME i
i 1 1 1 e . . .
Rm mmm HS SN
= Un c Un =
Height Difference in Metres
Fig. 6 Histogram of Height Differences
Red - Green
2
=
©
a
Lei
e
S
eg
ST<0 [m=
0’Z<S'I
0°Z 240qy mm : C s -]
= »
= =
®
S v
m po
> un
Height Difference in Metres
1 1 e [—]
- of z
V v eV Vv
1 e e m
Un
Fig. 7 Histogram of Height Differences
Blue - Red
8. ACKNOWLEDGEMENTS
The authors would like to acknowledge the support of the
British Geological Survey (BGS), in particular Dr. D.
Greenbaum and Dr. D. Tragheim, the National Remote
Sensing Centre (NRSC) Airphoto Group, in particular Mr.
M. Holt, and the Natural Environmental Research Council
(NERC). The views expressed in this paper are those of
the authors which are not necessarily those of BGS,
NRSC or NERC.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
