3 RESULTS AND DISCUSSION
The MVS workflow was applied to the terrestrial and the UAV
image datasets. For the UAV-MVS dataset, 151 of 153 images
chosen were processed resulting in a point cloud +175 m by
~60 m containing ~7.3 million points (~1-3 points per cm”). For
the T-MVS dataset, 174 of 179 images chosen were processed re-
sulting in a point cloud ~175 m by ~60 m containing ~6.3 million
points (33-5 points per cm”). Screen shots of these two clouds
and close up views of two 1 m staves are shown in Figure 3.
(c) The close up view of the UAV-MVS point cloud (point
size = 2).
(d) The close up view of the T-MVS point cloud (point
size = 1).
Figure 5: The derived point clouds.
Both point clouds have sparse sections in the woody scrub bush,
dead bushes and longer grasses. The UAV-MVS dataset has more
points representing vegetation in the central portion of the focus
area, this is not surprising due to the occlusion caused by tak-
ing the T-MVS photography from the water side of these bushes.
Both point clouds have a high density of points on the erosion
scarp and for soil and rock in general, even where the scarp is
overhanging. The texture of the ground in these areas is ideal for
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
feature identification as there is a lot of rocky gravel and shell grit
in the soil and the beach is very pebbly.
To analyse the effect of surface composition on point density pro-
files were visualised and compared. For illustrative purposes a
number of screen shots are provided that show regions or views
of interest. The Eonfusion scene is a far better viewing environ-
ment than the flat screen shots provided as the view perspective
can easily be adjusted to focus on interesting features from vari-
ous angles.
As can be seen in Figure 6(a) and Figure 7(a) showing profile
A the blue UAV-MVS points are amongst or slightly below the
T-MVS points. As the profile crosses the vegetation the sparse T-
MVS points on the occluded side of the bush can be seen amongst
the relatively dense UAV-MVS points.
25
% UAV-MVS
3 ST <
i
E :5 A
=
E
9 i
2 i
E Y
52626335 52626345 52626355 52626365 .
GDA94 Zone 55 Northing (mj
(a) Profile A.
VAV-MVS
2s T-MVS
[c pA bt
maa
Height (m)
Q h $ m 0 bÉ M5 gom
5262637
u
tas
o
tas
on
w
^
5262635 5262636
GDA94 Zone 55 Northing {m}
(b) Profile B.
Figure 6: A 1 cm wide profiles of the UAV-MVS and T-MVS
point clouds.
On the pebbly beach the UAV-MVS cloud is consistently below
the T-MVS cloud («1 cm) (see Figure 6(b) and Figure 7(b)). This
may simply be due to differences in the Helmert transformation.
Coregistration would be required to assess this further in a future
study.
The Poisson surfaces derived from these two clouds produced
new point clouds of surface vertices. The UAV-MVS Poisson sur-
face point cloud (referred to as UAV-MVS Poisson) has 2.3 mil-
lion vertices and the T-MVS Poisson surface point clouds (re-
ferred to as T-MVS Poisson) has 1.8 million vertices. After clean-
ing, the number of vertices were reduced by ~1100 and ~6000
points respectively. To visualise and qualitatively assess the ef-
fectiveness of the Poisson reconstruction and compare it to the
raw point cloud vertices (which would be used to create a dense
triangulated mesh surface), the extracted profiles were overlaid
and visualised in Eonfusion.
