PhD Lecture
by
Henrik
Rojas Nagel
Friday
17th June 2005, at 13.00
This thesis is based on Henrik
Rojas Nagel’s
research work at:
Department
of Health Science and Technology
Associate
Professor Simeon J. Simoff, University of
Technology Sidney
Professor
Dr. Bernd Fröhlich, Bauhaus-Universität Weimar
Associate
Professor Claus B. Madsen, Aalborg University
Abstract
In
this thesis is explored how Visual Data Mining (VDM) can be enhanced
with
Virtual Reality (VR). VR lets users explore Virtual Worlds (VW's)
from the inside-out by allowing them to continuously navigate to new
positions
inside these, in order to obtain more information from them. The VW's are, in this thesis, based on 3D scatter
plots, where
objects with attributes (glyphs) are used as markers. To solve the
problem of
choosing statistical variables from a given dataset for mapping to
glyph
attributes, a three step strategy with gradual selection of interesting
combinations of variables is proposed.
In
this thesis, the general approach for designing glyphs is to make use
of the
multi-modal audio-visual aspects of VR to create attributes that
perceptually
are distinguishable from each other. While there are multiple
possibilities
available for extending the visual attributes of ordinary glyphs, such
as by
giving them sound attributes, this thesis focuses on the visualisation
of arbitrary temporal developments.
To
perform experiments in VR, a general-purpose VR software framework
called VR++
has been developed, which supports VR visualisation, interaction,
encapsulation, modularity, inter-disciplinarity
and
distributed computing. It is designed to inherently support multiple,
simultaneous projects and multiple, simultaneous developers. Each VR++
developer adds modules to a large pool of such, and a program is
created by
specifying which modules are to be used, on which computer they should
be
executed, and how they should be connected. New projects can thereby
start with
reusing some of the available modules, and only have to create their
own
modules for functionality that is missing. VR++ automatically sends
data
between these modules and multiple communication protocols are
available.
On
top of this, a VDM software framework called 3DVDM has been developed.
In
3DVDM, some of the possible new temporal data exploration methods,
based on
real-time calculation of arbitrary temporal developments, have been
implemented. One of the explored approaches is to enhance glyphs with
motion
attributes. Another is to visualise continuous streams of data by
continuously
deleting and recreating the graphics in 3D scatter plots, as new data
arrive.
To create a clear temporal development, the visualised data is sorted
according
to one of its statistical variables, so that the sort variable, in
effect, is
mapped to user time.