Homepage of
Ben Daubney
Contact Information:
Department
of Computer Science
Swansea University
Singleton Park
Swansea SA2 8PP
United Kingdom
Email: B.Daubney@swan.ac.uk
Twitter: @BenDaubney_CV
Current Research
I am shortly about to start a position here
at Swansea developing algorithms for real-time interactive segmentation of
medical volume data. The purpose of the work is to develop segmentation tools
that are of use to medical practitioners and allow them greater control over
how the segmentation is performed, this is in contrast to the current drive in
Computer Vision to fully automate the entire process. The work is funded by National Institute for Social Care and Health Research (NISCHR) and is a partnership
between the Research Institute of Visual Computing (RIVIC), the NHS in Wales and Aberystwyth, Bangor,
Cardiff and Swansea Universities. The project page can be found here. I will be continuing
to work on the project with Xianghua Xie.
I currently work as a Research Officer at
Swansea University where I am funded as part of the Research Institute of Visual Computing,
which is a collaboration between several universities across Wales. My current
supervisor is Xianghua Xie and our work focuses
on how to model and represent greater uncertainty in vision problems. My
feeling is that often we are forced to extract more information from weak
observations than is really available, the result is that we often draw the
wrong conclusion. For example in 3D pose estimation we will often attempt to
extract a single solution when perhaps the available evidence (e.g. a binary
silhouette) is ambiguous and not strong enough to achieve this. In these
situations it is surely better to represent all possible solutions that an
observation can explain until a time when more information becomes available.
This raises two important questions: Firstly, how do we model uncertainty over
a much larger volume of the solution space than is currently possible?
Secondly, how do we update this model given a new set of observations without
it becoming computationally inhibitive? It is these questions we are hoping to
tackle in our current work.
About Me
I first undertook my undergraduate studies at Cardiff University
where I gained a BSc in Astrophysics, following which I worked for a short
period in the Astronomy
Instrumentation Group based at the university. Deciding to take a change in direction I
accepted a PhD studentship, funded under the EPSRC grant "Understanding
Biological Motion using Moving Light Displays", at the University of
Bristol studying the use of motion to extract 3D pose from a sequence of
images. My supervisors on this project were Neill Campbell and David Gibson, I passed my viva and
was awarded a PhD in November 2009.
Interests
Originally I grew up in the small North Devon village
of Woolacombe, this is a popular surfing village and as such am quite a keen
surfer. During my undergraduate degree and PhD I found less and less
opportunity to go surfing, however, since moving to Swansea, which has some of
the best surfing beaches in the UK, I have started to regularly get back in the
water again. There are a few pictures etc on the following page Surfing. I also enjoy cycling and am a member of a
local young walker club the Tawe
Trekkers.
Publications
Estimating Pose of Articulated Objects using
Low-Level Motion
Ben Daubney, David Gibson and Neill Campbell,
to appear in a special issue of the Journal of Computer Vision and Image
Understanding (CVIU) on Semantic Understanding of Human Behaviour in Image
Sequence, early 2012.
pdf
Entropy Driven Hierarchical Search for 3D Human Pose
Estimation
Ben Daubney and Xianghua Xie, BMVC, 2011
(Oral acceptance rate 8%).
pdf
Tracking 3D Human Pose with Large Root Node
Uncertainty
Ben Daubney and Xianghua Xie, CVPR, 2011.
pdf videos
Star Proposal: Videos in Graphics and Visualization
R. Borgo, M. Chen, B. Daubney, E.
Grundy, H. Jänicke, Gunther Heidemann, M. Hoferlin, B. Hoferlin, D. Weiskopf,
X. Xie,
Eurographics, 2011.
Estimating 3D Pose
via Stochastic Search and Expectation Maximization
Ben Daubney
and Xianghua Xie, Conference on Articulated Motion and Deformable Objects
(AMDO), 2010. pdf
Estimating 3D Human
Pose from Single Images using Iterative Refinement of the Prior
Ben
Daubney and Xianghua Xie, ICPR, 2010. pdf poster
Using Low-Level
Motion for High-Level Vision
Ben Daubney, PhD Thesis, University of
Bristol, 2009. pdf
Monocular 3D Human
Pose Estimation using Sparse Motion Features
Ben
Daubney, David Gibson and Neill Campbell, IEEE Themis workshop - held in
conjunction with ICCV, 2009. pdf
Real-Time Pose
Estimation of Articulated Objects using Low-Level Motion
Ben
Daubney, David Gibson and Neill Campbell, CVPR, 2008. pdf poster
Sequences:
fig5a, fig5e, fig6, fig7a, fig7e.
Estimating Gait Phase
using Low-Level Motion
Ben Daubney,
David Gibson and Neill Campbell, IEEE Workshop on Motion and Video Computing
(WMVC) - IEEE Winter Vision Meetings, 2008. pdf
Using Low-Level
Motion to Estimate Gait Phase
Ben
Daubney, David Gibson and Neill Campbell, International Conference on Computer
Vision Theory and Applications (VISAPP), 2008.
Extra Stuff
November 2010 - BVI Seminar: Invited
to present current research to the Bristol Vision Institute.
Seminar title " How to Robustly Track People using Weak Visual Cues".
April 2010 - RIVIC Graduate School: Part of
the organizing committee for the RIVIC
2010 Graduate School held at Swansea University.
March 2010 - Discover! Workshop: This
was an outreach workshop for 12 to 13 year old girls from the local area to
promote and encourage woman into Science and Engineering. As well as helping on
the day I gave a talk "Computer Vision: Getting Computers to Understand
the Visible World".
October 2009 - BMVA Meeting on Articulated Motion: Presented my work on using Low-level Motion for Pose Estimation. Talk Summary
is here
Links
KLT implementation - Stan Birchfield's C implementation of the KLT feature tracker.
CImg Library - A C++ library used to display/manipulate
images.
Netlab - A Matlab toolbox "designed to provide the central tools necessary
for the simulation of theoretically well founded neural network algorithms and
related models."
(31/3/2011)