ographs. It
f the DSM
calibration
ind the co-
as. The CO-
carried out,
s from the
h that the
is used to
1m, which
ches of the
automatic
1 in which
ignored in
lity of the
With only
achieved
the points
gineering
sampling
/ we have
del. If the
the value
the three
SM. This
ound and
> original
step, the
| method,
the area,
? process
1 will be
process.
Is on the
with:
> volume
Boersma, Saskia M.
44 Experiments
Different experiments were performed to review the achievable accuracy of the three-camera system. The first
experiment consisted of a flat surface with texture. From this flat surface a DSM was generated. DSMs were also
generated from plastic models of different sizes of pressure sores. These plastic models lack texture. We found that the
lack of texture caused difficulties during the DSM matching process. Texture projection was added in the final
experiments to review the improvements in the accuracy. The plastic wounds were also imaged again with texture
projection. It turned out that the accuracy improved with the use of texture projection.
Figure 7 shows the created DSM for wound number 1. The figure shows that blunders are filtered out and that the
resulting DSM is a smooth surface. Softplotter gives an indication of the theoretical precision of the generated DSM.
The theoretical precision of the generated DSMs is at least 0.5 pixel. This corresponds, with a pixelsize of 0.0074 mm
and an average distance of 30 cm, with a precision of 0.2 mm in object space.
This result seems to be too optimistic for the precision of the generated DSMs. The different models of the same objects
are compared to get more insight in the achieved precision of the DSM. Numerical results of the experiments are
summarized in table 2. The average, standard deviation and maximum value of the differences between the three
models is indicated for all test objects, with and without texture projection. In most cases we found a normal
distribution for the differences between two models.
Object Difference model Difference model Difference model
1/2 — 2/3 1/3 — 2/3 1/2 — 1/3
(in mm) Avg. Stdv. Max. Avg. Stdv. Max. Avg. Stdv. Max.
Flat surface (texture) 0.329 | 0.398 | 1324) 0230 0456 2320! 0336 0,578 | 2423
Plastic wound 1 (no texture) 0.105 | 0.942 | 10.050 | 0.238 | 0.891 | 12.034 | -0.121 | 0.990 | 16.579
Plastic wound 2 (no texture) 0.130 | 1.178 | 19.332 | 0.220 1.055] 11175 -0.326 | 1307 11463
Plastic wound 3 (no texture) 0.194 | 1.163] 14.377 | 0.231 | 0.985 | 8.397 1] 0,195] 1.249] 7.614
Head wound (no texture) 0.049 1.188 | 8.405 | 0.078 1.018 | 8.814 | 0.030 | 0.971 | 11.220
Pressure sore (no texture) 0.240 | 1.499 | 7.048 | -0.081 1.535 | 11.643 | 0.488 | 2.742 | 14.095
Pressure sore (texture) 0.279 | 0.753 4.210 | 0.327 | 0.676 | 5.174 | -0.080 | 0.803 8.421
Table 2: Accuracy of created DSMs
The table shows that the repeatability that can be achieved for a flat surface with texture projection is about 0.5 mm in
height. The repeatability that can be achieved for the plastic models of pressure sores, without texture projection, is
better than 1 mm in height. Texture projection should improve the results. The precision that can be achieved for a
pressure sore with texture projection is 0.7 mm in height.
Nexen Final result
ih
Lu
i
Hi
| is 3 e ap
i X tt j|
b 4 1 T
UE à
Diae Va
Ye)
4
i -2
P 5 uu
Ie t i} ea
Pa o ZA
É I eene s y 24
3 3 SL 5 Ÿ i Re :
3.3, ! 6 i 22
(Scan 2 >
Yim) 20
Ty 10
d
Ws
it
Bo
D NND SX B 50. ~ "10
; Y »- — yo (8 et Y (cm)
8) 35 8
lbi X (cm)
Figure 7: DSM of plastic wound 1
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 89