Underwater photogrammetry
HN 116 — 06/2020
29
resulting reduced point cloud of MBES and the
photogrammetric point cloud were then analysed
for the surface density. This measure calculates
the amount of points In a given neighbourhood
to estimated planes of a radius R and extrapolates
this number to 1 m 2 R was chosen to 20 cm, I.e.
approximately 2 x GSD of the MBES data. As the
Investigated structure was mostly flat, this metric
was chosen over the volumetric density. Tabje.3
shows the number of points and densities next
to the total area covered by the given data set. It
Is worth mentioning that the covered area Is es
timated by projecting the point clouds to a plane
and then calculating the area of the enclosing
polygon. Thus, the amount of points cannot be
simply divided by the covered area to obtain the
mean surface density.
The table prominently shows the complemen
tary characteristics of the two sensors. Where, on
the one hand, the photogrammetric data had
a high density with several hundred thousand
points per sguare metre, the MBES provided a
very much higher coverage of the area In a com
parable amount of time needed for each of the
methods.
4 Summary and outlook
The purpose of this feasibility study was to fuse
MBES data with 3D point clouds derived from
photogrammetric Image triangularon In order to
Improve data density. The paper presents a pro
cessing chain for combining hybrid point clouds
generated from photogrammetric Imagery and
hydroacoustic systems. MBES data hereby posed
an Initial starting point thanks to Its good abso
lute positioning accuracy and thus was used to
georeference and scale all further data. The point
cloud from Imagery was then to be transformed to
best match the MBES data via ICP.
The resulting data set provides an objective rep
resentation of a wreck, suitable for Improved deci
sion-making concerning underwater obstructions.
It shows good congruence of MBES data with the
photogrammetric point cloud. Even though, apart
from the MBES data, no Independent scale could
be Included for this data set, It Is observable that
the deviations between the two data sets achieves
an expected level of precision of 5.3 cm. Major
deviations result mostly In areas that are hard to
ensonlfy by the MBES (I.e. occluded areas near the
seafloor) and on the edge of the photogrammetric
point cloud with probably fewer measurement
samples and a worse acgulsltlon geometry. The
surface density Is highly Improved by several or
ders of magnitude due to the Imagery-derived
point cloud.
For point cloud registration, Initial values were
obtained by manually selecting salient points
such as the bow In the Imagery. Determining
correspondence between the salient points In
I Data set
Number of points
Area covered [m 2 ]
Surface density [pts/m 2 ] 1
MBES full
160,661
498
255
MBES closest points
3,791
22
158
Photogrammetry
12,324,258
22
535,069
Table 3: Point density metrics ofthe photogrammetric point cloud and MBES data
both point clouds was gulte challenging due to
missing local geometric gradients on the wreck's
starboard side. Automatic detection by suitable
feature detectors or marked positions should sig
nificantly Improve this processing step and avoid
manual Interaction In future works. Furthermore,
colour Information was Included by this fusion
using Image enhancement as a preprocessing
step. With this method, a green colour cast was
removed and a more natural representation of
the wreck was thus achieved. Though processing
times of photogrammetric data may go up to sev
eral hours or even days with large data sets, many
processing steps can be automated to a certain
degree.
During the underwater Image acgulsltlon, sev
eral practical problems occurred which had to be
taken Into account when performing observations
from a ROV and Integrating a camera system. The
tether connected to the ROV was often tangling
around edges of the wreck, thus limiting manoeu
vrability. Furthermore, the cable of the camera was
a regular Ethernet cable that had not been espe
cially ruggedlsed for high sea applications and was
rather short (70 m). This resulted In only parts of
the starboard side of the wreck being observed.
Future developments will lead to Improvements
on these Issues. Since turbid water conditions re-
gulred short acgulsltlon distances of approximate
ly 1 m, the remote control was a challenging task
as well. Currents, tides and tether entanglement
often Interfered with the operating commands
and thus affected the movement of the ROV.
Therefore, practice and careful manoeuvring dur
ing such challenging acgulsltlons are Imperative.
The presented workflow Is not restricted to a
ROV-borne data acgulsltlon. Continuous develop
ments In sensor technology leads to miniaturised
sensors that allow diver-based applications, too. In
this context, the Integration of additional sensors
on the measurement platform providing com
plementa! Information (such as sonar and Inertial
measurement unit) Is also part of our future work.
It can be concluded that the photogrammetric
method provides a higher grade of detail and ac
curacy, even when observing geometrically chal
lenging structures, such as a sidewall providing
only very low geometric Information In line of
sight, posing a dead reckoning problem. Using
Image enhancement, significant contrast Increase
could be achieved, resulting In more Images being
aligned and higher coverage ofthe wreck. This ap-