Underwater photogrammetry
28
Hydrographische Nachrichten
Fig. 7: Fused point clouds by photogrammetric analysis (RGB) and MBES soundings (coloured points) in side view (left) and isometric view (right).
The dense point cloud filled areas lacking MBES soundings due to occlusion, such as on the bottom of the wreck. Colour scale depicts depth in metres
code to the ground control points. After coarse
alignment, fine registration was performed using
the Iterative Closest Point (ICP) algorithm (Besl and
McKay 1992). The result Is a georeferenced point
cloud In ETRS89/UTM coordinates, which can be
directly Integrated In the following data process
ing chain of BSH hydrographic surveying. The hy
brid origin of the data Includes furthermore RGB
colour values and MBES backscatter values that
can be attributed to the points. Additionally, the
photogrammetric Imagery can be transformed
to the relevant terrestrial reference system and
stored, as well, providing a georeferenced Image
database for further Information on the wreck.
3 Results
Fig.,.7 shows an overlayed photogrammetric col
oured point cloud with the MBES data. The data
depicts the starboard side of the wreck's bow. The
area covered by the photogrammetric point cloud
Is about 13 m x 2 m, which Is approximately half
of the wreck's length and almost the entire height
above the seafloor on the starboard side.
It Is clearly visible that the photogrammetric
point cloud has a significantly higher point den
sity and provides colour Information. Furthermore,
occluded areas near the bottom of the wreck can
be observed with the photogrammetric data and
thus provides completion of the data In these ar
eas. This Is observable In Fi_g._7 (left), where areas
of the wreck's bottom are filled with points In ar
eas that were occluded to the MBES and thus not
covered entirely. For accuracy evaluation, cloud-
to-cloud distances were calculated, using least
sguares planes on the MBES data (Low 2004). Dis
tances from the photogrammetric point cloud to
local planes of the MBES data are calculated and
shown In Fig_._8.The mean cloud-to-cloud distance
Is 0.053 m. Though not fully representative of the
absolute accuracy achieved by this method, these
numbers representa measure of Internal accuracy
for the combination of both data sets. Apart from
that, higher deviations occur In the lower part of
the point cloud In areas that are not as densely
covered as the rest of the point cloud.
Comparing point densities, the MBES point
cloud was reduced for the subset with the closest
points to the photogrammetric point cloud. The
0.200 rn
0.150
0.100
Fig. 8: Cloud-to-cloud distances referenced to least squares planes on MBES point cloud (scale unit in metres)