XXIX-B8, 2012
| matrix
ection center
1 ground coordinate
stereo image for 3D
orporated. With this
very less knowledge
use this system.
ivigation seamlessly
th) is also integrated
e of the real world
ncludes the map, or
te project which will
ISers position with
function has been
olume directly from
ulation is a function
number of trees per
of timber volume
volume
DY
areas, UltraCamX
red with GSD of
performed during
ess of the forest.
region Japan. The
containing planted
ly consisting oak,
can also be visited
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
htp;//fddb.ffpri-108.affrc.go.jp/en/03 ogawa/index.html
The area used for measuring the tree heights for this
experimental study is shown Figure 3. The area contains
conifers trees, deciduous trees, and the ground surface (fields,
roads).
33 Outline of the experimental study
The tree height, used in the calculation of timber volume, in the
system is calculated by subtracting the elevation of the crown
top and the ground level. Hence, the accuracy of the
measurement depends on the accurate measurement of the
crown top and the ground surface. The top of the tree crown is
measured directly on the system by putting the cursor at the top
of the tree and by adjusting the tree height within the stereo pair.
To measure tree height, the ground elevation under the tree
canopy obtained from Triangulated Irregular Network (TIN)
constructed from the total station ground survey points was used.
The difference in elevation between the tree height measured
and the ground height under the tree canopy measured from the
TIN gave the tree height.
To evaluate the accuracy of the tree height measurement, the
measurement of the elevation of the top of the tree crown
obtained by the automatic stereo matching between the image
pair was used as the base data for this experiment.
The ID numbers in Figure. 3 are the measurement points of
confiers trees, decidious trees and the ground points.
For the accuracy evaluation of ground height measured by
Morittai, the ground elevation measured by Morittai was
compared against ground elevation constructed from LiDAR
data.
3.4 Result and Discussion
As a test of the system, two persons were assigned to measure
the top of the conifer trees, deciduous trees and the ground
elevation. One of the person employed for the test was a
beginner with no prior experience in forestry and photo
interpretation. However, the other person who took part in the
experiment had a lot of experience in photo interpretation but
no knowledge about forestry. The accuracy of the height
measurement in the Morritai by the participants is shown in
Figure 4. In the figure, ID number 1-9 represents measurement
of conifer tree heights, ID 10 to 18 and 25 to 40 represents
measurement of deciduous tree height and ID 19 to 24
represents the measurement of ground height. From the figure,
the general trend shows that the beginner generally
underestimate the height measurement. However, the
experienced person has slightly overestimated the height
measurement at places. It is not clear whether it is an empirical
thing or due to individual differences, and whether the
differences could be corrected by systematic practice.
Figure 3. Experiment area
"tipos yx poricnecd
—<Baginner
Difteronce
1383779111215171921232027 283155 353138
ID number of points
Figure 4. Comparison of accuracy evaluation of height
measurement
Table 1 shows the accuracy evaluation of the test cases. The
overall error of the height measurement including both type of
trees and ground height was about 1 m for experienced and
about 3 m in case of beginner. Hence an average error obtained
by considering observations from the both experienced and the
beginner observer was about 2 m.
Experienced Beginner
SD (m) RMS (m) SD (m) RMS (m)
: 1.5 1.7
Conifers 0.6 2.9) 0.6 (2.5)
x 1.3 37
Deciduous 1.3 (24) 2.4 (-12.0)
0.6 3.9
Ground 0.6 (-0.9) 1.8 (54)
Total 12 1.3 2:1 3.4
Table 1. Accuracy results of height measurement
From the table it is seen that the person with the prior
experience has measured with lower error level. The result
shows that for the beginner, measuring tree height for conifer
trees was much easier than for the deciduous trees. The top of a
conifer tree is easy to be distinguished for the beginner
compared to the convex shape of the deciduous trees. Hence
beginner had less trouble deciding the tree top. But the tree
shape was little of a concern for the person with prior
experience. The interesting observation from the experimental
study shows that the person with prior experience could
measure the ground height with the highest accuracy followed
by deciduous tree and conifer trees. However, considering the
beginner, it was just opposite. It seems that the beginner will
have an ease in measuring the structure with distinct shape than
plane like ground or convex shape like deciduous trees.
The maximum difference in height measurement observed by
the person with a prior experience was 2.9m for a conifer tree, -
2.4m for a deciduous tree and -0.9m at a ground surface. In case
of the beginner, the maximum difference in the height
measurement for a conifer tree was -2.5m, a deciduous tree was
-12.0m and for a ground surface was -5.4m. The maximum
difference of -12.0m height was observed at point ID 12. The
location of ID 12 is depicted in Figure 3. This was an
exceptional error produced due to the carelessness during
observation made by the beginner. In general, the experiment
shows that the maximum error in tree height measurement by an
