In the first stage, with the aerial photographs covering the
surveyed area as the input data, our stereo processing system
produces a dense digital surface model (DSM) of the area.
Based on this, the digital terrain model (DTM) or bare-earth is
estimated based on the DSM, ortho-rectified colour image, and
ortho-rectified infra-red image through our novel DEM
estimation algorithm. Then we obtain the average tree height of
the area by comparing the DSM and DEM, thus acquire the tree
height map for the forest management GIS system.
Then the proposed GIS system takes the tree height map as its
basic input data. It is possible to display the following data in
the proposed GIS system, such as DSM, orthophoto and forest
base map that includes the area boundary data, tree species,
land owner and etc. The proposed GIS system also provides
interactive interface to estimate the stand density of tree and
then further to get forest resource based on the stand density
and the tree height map. Users of the system can easily estimate
the forest resources such as the volume of lumbers or that of
lumber remnants along newly built road. The estimation of the
forest resources helps to assure the stable supply of lumbers and
the effective utilization of forest resources without wastes.
3.1 Image Analysis of Aerial Photograph for Forest
Resource Estimation
The information of tree species, height and density is necessary
for forest resource estimation.
3.1.1 Tree Height Estimation using Aerial Stereo Photos
It is possible to estimate tree height using aerial photo by
subtracting the height of the ground (DTM) from the height of
the top of the tree (DSM) in the forest.
We have developed a stereo processing system (Koizumi, 2009).
The system, mainly applied to urban area, can generate DSM
including building height pixel by pixel using stereo matching
from plural aerial photographs. That is, in the case of 20cm
ground resolution of aerial photograph, DSM is generated on
every pixel in the same resolution of 20cm. But in forest area,
DSM doesn’t necessarily have comparable accuracy to in urban
area for the following special factors in the forest images, for
example, existence of texture, complex shape shadow, different
contrast caused by various direction of slopes, steep height
change similar to buildings.
Then, we analyze the reason of the deterioration of the
matching accuracy and further improve the stereo processing
system to deal with geographical steep undulations in forest and
also reduce the error of relative orientation on stereo images
due to the error of aerial triangulation. As a result, compared
with conventional system, for the input forest area data having
y-parallax, the matching noise in DSM caused by geographical
steep undulations and y-parallax is reduced in the improved
system (figure 2).
(a) 3D model by conventional system (b) 3D model by improved system
Figure 2. Concept Image of DTM Estimation
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
Next, we estimate the tree height in forest based on the DSM,
ortho-rectified colour image and ortho-rectified infra-red image
generated from aerial stereo photographs (Wang, 2011). At first,
bare-earth regions such as road and vacant land in the
surroundings or inside of the forest are automatically extracted
from images by judging vegetation using Normalized
Difference Vegetation Index (NDVI, Equation(1)) from ortho-
rectified colour and infra-red images. Then, tree height is
estimated from the difference between the heights of extracted
bare-earth region and the surrounding tree regions. At last,
height of trees inside the forest is estimated from the formerly
obtained tree height of the trees near bare-earth region as a clue
(Figure 3).
NIR — RED
NOI MRCHRED. ()
NIR + RED
where NIR = the spectral reflectance measurements
acquired in the near-infrared channel
RED - the spectral reflectance measurements
acquired in the red channel
known known
unknown
Height of ground
(known)
Estimation
Height of ground
in forest (unknown)
Figure 3. Concept Diagram of DTM Estimation
For the accuracy evaluation of DSM and DTM obtained from
aerial images, we compare them with DSM and DTM generated
from LIDAR, and also with DSM and DTM generated by
stereoscopic vision in the Mie University experiment forest
(Figure 4, Figure 5). The result of evaluation shows that the
average difference of DSM with that from LIDAR is almost
zero and the root mean square error is 4.8 meter. And the
average difference of DTM with that from LIDAR is less than
one meter and that accuracy is almost same as stereoscopic
vision. Considering the cost, our system provides a much
cheaper way than LIDAR to acquire the same ground resolution
level of data. And the aerial photographs can also provide
information of tree species and information necessary for
density estimation. We conclude that DSM and DTM generated
from aerial photograph provide enough accuracy for practical
purposes of forest resource management.
Aerial Photograph Generated DTM
Figure 4. Result of DTM Estimation
3.1.2
It is
resou
plant
cedar
SUITO
com
fores
exist:
frequ
upda
these
not u
Ther
estin
cond
in T
spec
of tr
shov
valu
Next
phot
anal;
betw
Average value of Blue
3.2
As |
heig
for
