| Q&A |
 ACVT
Anton van den Hengel
Oct. 2007 |
| |
 |
"So
VideoTrace is designed to work with whatever video you
need to grab an object from, rather than specifically
shot image sets. [...] This means that you can create
an accurate model from a video in minutes with VideoTrace,
whereas using PhotoModeller requires that you provide
much more information manually, making it a much more
arduous process. "
<
VideoTrace |
|
| |
|
| Q1 |
Can
you introduce the VideoTrace project and its purpose? |
| A1 |
VideoTrace is a system for interactively generating realistic
3D models of objects from video—models that might be inserted
into a video game, a simulation environment, or another video
sequence. The user interacts with VideoTrace by tracing the
shape of the object to be modelled over one or more frames of
the video. By interpreting the sketch drawn by the user in light
of 3D information obtained from computer vision techniques,
a small number of simple 2D interactions can be used to generate
a realistic 3D model. Each of the sketching operations in VideoTrace
provides an intuitive and powerful means of modelling shape
from video, and executes quickly enough to be used interactively.
Immediate feedback allows the user to model rapidly those parts
of the scene which are of interest and to the level of detail
required. The combination of automated and manual reconstruction
allows VideoTrace to model parts of the scene not visible, and
to succeed in cases where purely automated approaches would
fail. |
| |
|
| Q2 |
What
is the difference between VideoTrace and existing 3D image-based
reconstruction tools such as Realviz ImageModeler? |
| A2 |
VideoTrace
allows the user to model arbitrary objects in normal video.
Pretty much any video that you can apply a camera tracker to
you can use as input to VideoTrace. That makes it a lot more
flexible than something like ImageModeller which requires that
you take a set of photographs of the object that you want to
model under a very controlled (and constrained) set of circumstances.
So VideoTrace is designed to work with whatever video you need
to grab an object from, rather than specifically shot image
sets. There are some similarities with PhotoModeler, but the
interaction in VideoTrace is much more intuitive, and more powerful.
This means that you can create an accurate model from a video
in minutes with VideoTrace, whereas using PhotoModeller requires
that you provide much more information manually, making it a
much more arduous process.
The flexibility and power of the VideoTrace modelling process
means that it can be used to generate models for all of the
purposes you might find, or per-pixel depth maps for compositing,
from whatever video you need to use it on. No special cameras
are needed, no laser scans, no tape measures, just a simple
intuitive tracing process. |
| |
|
| Q3 |
The
VideoTrace demo is very impressive. How long has the system
been developed? |
| A3 |
The
system has been in active development for 2 years, but it builds
on work that has been carried out by the group over more than
10 years in the area. We've just (as in minutes ago) released
the first beta version to a limited set of testers, so it's
certainly a very usable system, which we're hoping to keep developing
for a few years to come. |
| |
|
| Q4 |
Such an application must be very CPU-hungry. Can you tell us
the hardware required to run VideoTrace? Does its architecture
take advantage of multi-core processors? |
| A4 |
It's
certainly not going to run on your cell phone any time in the
near future, but you really don't need to have all that powerful
a machine to use it. Most of the work I've done with it has
been on my laptop, which really isn't anything special. All
you really need is a graphics card with OpenGL support, and
it doesn't even need to be the latest version. |
| |
|
| Q5 |
In
which format(s) does VideoTrace export the 3D models? |
| A5 |
VideoTrace
exports in a few formats, but the most useful one is VRML. Most
packages import VRML, and there are translators for pretty much
every other relevant format that you can think of. The current
Beta has a limited set of import formats, but the next version
will import from more of the camera trackers. We've implemented
the functionality already, it's just a question of documenting
it really. |
| |
|
| Q6 |
How
did researchers from The
Australian Centre for Visual Technologies and The Oxford
Brookes Computer Vision Group meet around this project? |
| A6 |
We've
known each other for a surprisingly long time, but really
the primary motivator was that Phil Torr (from Oxford Brookes)
partly supervised the PhD of Anthony Dick (from Adelaide).
The collaboration has been extremely positive, and one for
which we've just got and other 3 years research funding. |
| |
|
| Q7 |
Does
VideoTrace make use of existing software libraries ? Can you
tell us which ones? |
| A7 |
It
uses SSL and QT, but that's about it. |
| |
|
| Q8 |
VideoTrace
technology could provide a tremendous addition to existing 3D
modeling packages. Have you already been contacted by open source
or commercial software providers? |
| A8 |
Yes we're negotiating with quite a large number of companies
about the future of VideoTrace, but we haven't decided on anything
yet. It seems to take a while to get these things organised. |
| |
|
| Q9 |
What
improvements are to be done on the VideoTrace system? Can you
give us a brief roadmap? |
| A9 |
 We've
got a lot planned for improving the fidelity and flexibility
of VideoTrace over the next few years. We're looking initially
at using interactive dense matching (a technique from computer
vision) to improve the way we handle curved surfaces. We're
looking at how we might interactively de- and re-light objects
which are cut and paste between video sequences. We may look
at interactive camera tracking, there's a long list. |
| |
|
| |
|
| |
|
