International Archives of the Photogrammetry. Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
stereo - stereo - laser -
[m] laser image image
correlation correlation
Plot 22
Tops coniferous SD 0.79 3.83 4.17
n-9 Mean 0.36 5.07 4.71
Crown decid. SD 0.89 2.34 1.86
n=19 Mean 0.53 0.36 -0.16
Plot 50
Tops coniferous SD 1.57 1.18 1.79
n=9 Mean 2.33 2.01 -0.33
Crown conif. SD 5.79 7.78 7.64
n=16 Mean 3.56 -2.82 -6.38
Regeneration SD 0.88 1:33 0.86
n=8 Mean 1.83 0.53 -1.30
Ground surface SD 4.44 7.07 5.94
n=19 Mean -1.35 -7.67 -6.33
Plot 60
Tops deciduous SD 0.55 0.43 0.41
nzl7 Mean 0.64 0.17 -0.47
Crown decid. SD 0.76 0.35 0.63
n=ll Mean -0.18 -0.10 0.08
Crown edges SD 0.52 0.42 0.48
n=15 Mean -0.26 -0.25 0.01
Table 4: Height differences between the DSM's derived from
laser scanner (laser) and image correlation. and
stereo measurements.
laserscanner data since these measurements are at average
1.35m beneath the stereo measurements. Standard deviations
are 0.88m for the regeneration and 1.57m for the tree tops. For
the ground surface and edges of the crowns the standard
deviation reaches quite high values between 4.44 and 5.79m.
The comparison between the stereo measurements and the DSM
from the DMC data shows that the stereo measurements are
above the data derived by image correlation for the tree tops and
the regeneration. The mean values and the standard deviations
are comparable to the laser measurements. However, there are
huge differences of —2.82 and —7.67m in mean value for the
ground surface and for the crowns. Also, the standard deviation
reaches very large values with more than 7.0m for these classes.
The reasons for this are shown in Figure 5. The DSM derived
from the DMC data does not adequately represent the surface
structure of the forest. Especially, the valleys between the trees
are only poorly rendered. In addition, the tree tops are
underestimated. However, this effect is not as significant as in
the old growth stand.
Note that the smallest differences between the data types and
the lowest standard deviations can be found in the deciduous
stand. Comparing the stereo measurements and the laser data
the mean value for the tree tops is 0.64m, for the crowns -0.18m
and for the edges of the crowns -0.26m. The standard deviations
are between 0.52 and 0.76m for the crown edges and crown
measurements. The differences between stereo measurements
and the DSM derived form DMC data are even smaller. The
mean value ranges between -0.25 and 0.17cm for crown edges
and tree tops. All standard deviations are less than 0.42m. This
data also shows that tree tops are underestimated and valleys
between the trees are overestimated by the algorithm.
88
1265.0 e
|
plot 50 + fs 4
1260.0 | /
| " | yv NY |
| * Poem vi |
_ 12550 | ^o AN A / « / = Zl |
ë \ f f i
E | fF A NN) pa” |
v | | = \ \ { Y |
£ 1250.0 | / i : 1d i PX |
= ; j
v | / i ; i i
ë La) ] *
$ 12450 | ! 1
= | / |
= | \
2 If | i j 1
40,0 | / i j \ ® |
| i i 1 M
| ^ ! d à # |
1 y * \ i i
235,0 | á i
|
á |
1230,6 | - - ——— - ein - :
0,00 20,00 40,00 60,00 80,00 100,00 120.00 140,0(
distance [m]
plot 22 /
eight above sea ievel [m]
| ~~ Laser —d— DMC
902,0 |
0,00 20,09 40,00 60,00 80,00 100,00 120,00 140,00
distance [m]
Figure 5: Comparison of the laser scanner DSM and the DSM
derived from DMC images at typical vertical
profiles through the plots 50 and 22.
4.3 Delineation of individual trees
The results for the automated tree detection depend on the
resolution of the grid and on the data source. For the
laserscanner data it was possible to calculate a DSM with 0.5
and 1.0m resolution. This led to an increase in detection rate of
16 to 20%. Also the delineation of the crown shape was much
more accurate with a resolution of 0.5m. Comparing the results
for the spatial resolution of 1.0m achieved by the laserscanner
DSM and by the photogrammetric DSM the detection rate for
the laserscanner DSM was better than for the photogrammetric
DSM for plot 22 and plot 60 while the percentage of detected
trees were similar for plot 50. The reason for these results is the
relatively large distance between the trees in this plot. Their
crowns are separate, the canopy is not closed, so that tree
detection is easy even with a Im DSM which does not render
detected trees [%]
plot DSM data upper middle total
resolution — source layer layer
22 0.5 Laser 65.2 17.6 29.6
1.0 Laser 59.1 6.7 25.0
1.0 DMC 36.4 S7 14.9
50 0.5 Laser 86.0 66.7 84.8
1.0 Laser 65.1 33.3 63.3
1.0 DMC 67.4 0.0 63.3
60 0.5 Laser 54.8 0.0 37.8
1.0 Laser 38.7 0.0 26.7
1.0 DMC 21.9 0.0 15.6
Table 5: Results of the single tree delineation.
Internat
the surfac
DSM in pl
[t was prot
The result:
high resol
density of
better than
high point
The applie
matching
canopy su
scanner di
coniferous
and underc
mainly cat
erroneous]:
representin
cut by the
knowledge
feature poi
images. In
be observe
impossible
DTM gen«
results esp
matching r
been adopt
There are s
present al;
feature-bas
multiple ir
perfect ove
case the er
geometrica
would in tu
reliability
matching |
present wl
Furthermor
true 3D D$
knowledge
algorithm r
3D DSM re
tree model
trees are si
method. Si
image scale
These resul
images for
forested I:
application
Firstly, the
Secondly,
simultaneoi
informatior
of forests. '
inventory t;
and promisi
