nbul 2004
fi
a
N "
X
*
1
|
|
i
i
00 1,06
“DMC
‚00 0
1d the DSM
cal vertical
end on the
e.-For the
M with 0.5
tion rate of
> was much
' the results
aserscanner
ion rate for
grammetric
of detected
esults is the
plot. Their
o that tree
XX render
[7]
total
29.6
25.0
14.9
84.8
63.3
63.3
37.
26.7
15.6
oo
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
the surface perfect. The relatively bad results for the DCM
DSM in plot 60 are probably caused by the extend of the DSM.
It was probably to small for the tree finding algorithm.
S. DISCUSSION
The results show clearly the potential of DSM generation with
high resolution digital cameras for forest surfaces. The point
density of matched 3D points of more than 10 pts/m? is even
better than the TopoSys scanner, which is well known for its
high point density.
The applied method of DSM generation using a feature-based
matching approach could successfully reconstruct deciduous
canopy surfaces with almost the same accuracy as the laser
scanner did. However, the method failed especially when single
coniferous trees were present in the plot by cutting the tree tops
and underestimating the lower areas between the trees. This is
mainly caused by the following. Firstly, the matching works
erroneously in occlusion areas. Secondly, the single trees
representing objects with large x-parallaxes in the images are
cut by the algorithm and the matching parameters since no pre-
knowledge about the DSM is used. Thirdly, the matching of
feature points is error prone since it 1s restricted to two stereo
images. In some cases 30 % and even more mis-matches could
be observed which makes robust filtering of a surface almost
impossible. However, the procedure was originally designed for
DTM generation in open field areas. Therefore, the inferior
results especially in the plot 50 are only caused by the specific
matching method and the DSM reconstruction which has never
been adopted for tree reconstruction.
There are several ways to improve the matching strategy of the
present algorithm for canopy reconstruction. Basically the
feature-based matching part must take full advantage of
multiple image overlap. In our case we would even have a
perfect overlap of at least 6 images for the entire area. In this
case the error percentage would be drastically reduced because
geometrical constraints like epipolar lines could be used. This
would in turn lead to a significant improvement of accuracy and
reliability thanks to the multiple 3D intersection. Also, the
matching part must cope with occlusions which are always
present when perspective images of tall objects are used.
Furthermore, the simple 2.5 DSM model must be replaced by a
true 3D DSM representation based on a TIN structure. If pre-
knowledge about the terrain (e.g. laser DSM) is available the
algorithm must be able to use it. This approach would use a true
3D DSM representation to model the trees rather than a specific
tree model. The latter would also be possible once the single
trees are successfully delineated and located by an appropriate
method. Since the point density is mainly dependent on the
image scale a grid spacing of 50cm or even better is possible.
These results show the potential of DSM's derived from digital
images for forestry and other environmental applications in
forested landscapes. In comparison to laserscannning the
application of digital cameras shows two main advantages:
Firstly, the costs for the image acquisition are much lower.
Secondly, spectral information of high quality is collected
simultaneously. However, it is not possible to collect
information about the vertical structure and the ground surface
of forests. Therefore it is to be concluded that for future forest
inventory tasks a combination of both sensors is advantageous
and promising.
89
6. REFERENCES
Brandberg T., Walter F. 1999. An Algorithm for Delineation of
Individual Tree Crowns in High Spatial Resolution Aerial
Images using Curved Edge Segments at Multiple Scales.
Proceedings of Automated Interpretation of High Spatial
Resolution Digital Imagery for Forestry Proceedings of
Automated Interpretation of High Spatial Resolution Digital
Imagery for Forestry. British Columbia, Canada. .(eds. Hill,
D.A.and Leckie, D. G.) Natural Resources Canada, Canadian
Forest Service, Pacific Forestry Centre. Pp. 41-54.
Diedershagen O., Koch B., Weinacker H., Schütt C. 2003.
Combining Lidar- and Gis Data for the extraction of forest
inventory parameters. In: Hyyppá, Naesset, Olsson, Pahlen and
Reese, (eds.)Proceedings of the Scandlaser Scientific Workshop
on Airborne Laser Scanning of Forests. pp. 156-164.
Dralle K. 1997. Locating Trees by Digital Image Processing of
Aerial Photos. PhD thesis. Royal Veterinary and Agricultural
University. Frederiksberg. Denmark.
Gougeon F. A. 1999, Automatic Individual Tree Crown
Delineation Using a Valley-following Algorithm and a Rule-
Based System. Proceedings of Automated Interpretation of
Hight Spatial Rsolution Digital Imagery for Forestry Victoria,
British Columbia, Canada.(eds. Hill, D.A.and Leckie, D. G.)
Natural Resources Canada, Canadian Forest Service, Pacific
Forestry Centre. Pp. 11-23.
Hinz A., Dorstel C., Heier H. 2001. DMC - The digital sensor
technology of Z/I-Imaging. Photogrammetric week 2001.
Wichmann Verlag. Heidelberg. pp. 93-103.
Hese S., Lehmann F. 2000. Comparison of Digital Stand
Surface Models of HRSC-A and Laser Scanner for Forest Stand
Characteristics. IASPRS. Vol. XXXIII. Amsterdam 2000.
Heurich M., Schneider T., Kennel E. 2003. Laser Scanning for
Identification of Forest Structures in the Bavarian Forest
National Park. In: Hyyppà, Naesset, Olsson, Pahlen and Reese
(eds.) Proceedings of the Scandlaser Scientific Workshop on
Airborne Laser Scanning of Forests. S. 97-106.
Hildebrand G., Akca A., Djawadi K. 1974. Characterization of
the detailed roughness of natural surfaces by a digital surface
model. Proceedings: Symposium of Remote Sensing and
Photointerpretation. Banff. Alberta. Canada. 2: 755-765.
Krzystek P.. Wild D,. 1992. Experimental Accuracy Analysis
of Automatically Measured Digital Terrain Models. Robust
Computer Vision — Quality of Vision Algorithms. Wichmann
Verlag. Karlsruhe. pp. 371-390.
Wehr, A. and Lohr, U., 1999. Airborne laser scanning — an
introduction and overview. ISPRS Journal of Photogrammetry
and Remote sensing 54, 68-82.
Pretzsch H., Biber P., Dursky J. 2002. The single tree-based
stand Simulator SILVA: construction. application and
evaluation. Forest Ecology and Management. 162 (2002).
Persson A., Holmgren J., S6dermann U. 2002. Detecting and
Measuring Individual Trees using an Airborne Laser Scanner.
hotogrammetric Engineering & Remote Sensing. Vol. 68. No. 9.
September 2002. pp. 925-932.
