A, 2012
uve
;. elemo.ch).
WDR
red by the main
> field campaign
ire portrayed in
> 3D trajectories
ly analysed and
igation data has
'ssion of instan-
ition of missing
ies become par-
tion bathymetry
e portion of the
o prepare daily
nchronise such
ally provided as
) daily missions
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B4, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
e E EBwEBgm8EEIE
» aM enter
Figure 5: Digits (a) and masks (b) utilised by the digit classification
algorithm.
(26 from Mir 1 and 24 from Mir 2) for a total of 27h43 of video
recording. We processed each disk image using Apple's iMovie
software, extracting 50 audio/video quicktime files. These clips
were eventually uploaded to YouTube, as we considered it the most
practical and sophisticated tool to stream video for the project's
scope. In order to build the time reference string of each clip, given
the length of video data, we decided to use an Optical Character
Recognition (OCR) approach over video frames. Quicktime files
were further processed to obtain image sequences with a single
frame-per-second frequency, concordant with navigation data fre-
quency. OCR was carried out with a C++ script using OpenCV 2.3
(Open Source Computer Vision), a library of programming func-
tions for real time computer vision. The idea was to estimate each
video frame digit by comparing it with a known set of digits in the
form of an optimisation problem. Initially, from selected images
we created a sample set of 10 greyscale digits and a corresponding
set of 10 black and white digit masks (resp. pixel values 0 and 1).
Through code routines we compared the image's region of inter-
est corresponding to each one of the 6 time digits (hh:mm:ss) with
the sample set of digits and masks portrayed in Figure 5 (a) and
(b) respectively. Subsequently, the estimate of each desired digit is
obtained as a solution of an optimisation problem, which may be
formulated as follows
2 : 1 2
deren f IR Dj» Mili, ) ; (1)
where o denotes the element-wise, or Hadamard product, R rep-
resents a 24x26 pixel region of interest extracted from the origi-
nal image exemplified in Figure 4, while D; and M; are the ?-th
digit and mask, as depicted in Figure 5 (a) and (b) respectively.
The minimum result provides a numeric digit corresponding to the
one represented in the original video frame. Our code output a csv
(Comma Separated Values) file containing a numeric sequence of
digital timestamps. Due to strong changes in video frames back-
ground, which was varying from pale yellow to dark green, the
results were affected by errors. Hence, we used Excel and ad-hoc
algorithms to check for errors and correct them. The final result
was a time indexing string, like the one reported in the following
example, composed by substrings of initial time, ending time and
total seconds from the beginning, one for each continuous video
sequence inside a single mission video.
The resultant video index string is illustrated in Table 2 lines 33-36.
Using this strings, we were finally able to index videos on YouTube
and to synchronise them exactly with navigation data.
2.3 Existing WebGIS solutions
While developing our project-specific WebGIS platform, we have
carried out a comprehensive review of the Web-based GIS solutions
currently available. Specifically, the key features we were looking
for were the capability to manage dense time and space 4D datasets,
to be collaborative (i.e. multiple users being able to input and man-
age datasets), to be client-side, and to be easily configurable and
extensible. In particular, the application frameworks that have been
identified as potentially satisfying our requirements are MapFish
(www.mapfish.org), Geomajas (www .geomajas.org), and Map-
Bender (www .mapbender . org).
MapFish — A project of the Open Source Geospatial Foundation
(OSGeo Foundation, www. osgeo . org) that provides an open source
web mapping development framework that extends the Pylons Python
development framework with geospatial-specific functionality and
uses the Javascript libraries OpenLayers, ExtJS and GeoExt.
Geomajas — An open source webGIS framework with client-server
interaction that allows powerful displaying and editing of spatial
data. Like MapFish, Geomajas is an OSGeo project and is compli-
ant with Open Geospatial Consortium (OGC, www.opengeospatial.
org) standards (WMS, WFS, GML). Geomajas has a strong server
side focus where processing, styling, filtering, caching, etc. of
geospatial data happens, while a powerful data management and
visualisation interface is available for multiple users through their
browsers. Moreover, similarly to our approach, Geomajas is de-
signed with a plugin structure that allows extensibility and adapt-
ability.
MapBender — A back office software and client framework for
spatial data infrastructures. The software is implemented in PHP,
JavaScript and XML and dual licensed under GNU GPL and Sim-
plified BSD license. It provides a data model and web based in-
terfaces for displaying, navigating and querying OGC compliant
map services. There are many installations of MapBender, espe-
cially in Germany, some of which utilise the features of the jQuery
JavaScript framework (jquery.com) and displaying data profiles
coupled with planar mapping.
Subsequently, we have concluded that although theoretically possi-
ble, none of the considered frameworks has demonstrated the capa-
bility to facilitate the visualisation of synchronised multi-dimensional
data. Our dataset, like many coming from marine and aerial sur-
veys, has strong need of space and time data correlation. We have
therefor decided to develop a simple but powerful system to cater
for our specific usage scenario. Furthermore, we could not identify
an application which would provide a client-side implementation.
Client browser are becoming extremely powerful, as well as HTML
and Javascript languages. Moreover this feature allows a webGIS
system to run on mobile devices and be configured as mobileGIS
during surveying. We have concluded that this is a core requirement
in our field of application.
2.4 Developed Web-GIS platform
Correspondingly, the following items (a)-(k) provide a brief de-
scription of the Data Browser visual components as indicated in
Figure 6(a-k), respectively.
a) 2D Map - The central part of the data browser interface por-
trayed in Figure 6 is constituted by a Google maps-based interac-
tive map, which provides the geographical references of the col-
lected data. Google maps API was used to support the spatial part
of the survey's visualisation and cover the first two dimensions of
the data. The underwater trajectories of both submersibles Mir 1
and 2 are colour-coded in accordance to their correspondence to ei-
ther of the submersibles (red and blue for Mir 1 and 2 respectively).
When a trajectory is selected in the interface, related information is
displayed in the other parts of the page to display the most com-
plete overview possible of different aspects of this dive. Discrete
samples and observations are indicated with markers, as described
in the Visualisation Key portrayed in in Figure 6(b).
