Estimation of 3D Rigid Body Transformations from Image Sequences
Speaker: Stefan Lehmann, ITEE
When: 2004-05-12 14:00:00
Venue: 78-421
Host: Vaughan Clarkson
Abstract:The analysis of 3D rigid body transformations based on camera images
is one of the most challenging research areas in computer
vision. The accurate estimation of 3D rigid body transformations
builds the basis for image and object tracking
techniques. Application areas include teleconferencing, video
compression, robotics, camera calibration, surveyance and virtual/
augmented reality applications.
Common 3D motion estimation techniques can be loosely divided into
two classes. The first class relies on the computation of a sparse
set of feature correspondences. The second class depends on the
accurate computation of a dense optical flow field. Both of these
classes suffer from various drawbacks.
In this talk, a new motion estimation technique will be presented
that evaluates the spectra of two or more camera images. Our theory
bases on an integral projection model which has a correspondence to
standard parallel projection. A mathematical model of our theory and
the effect of rigid body transformation will be discussed. The
analysis of the transformation parameters with one camera only
results in an under-determined equation system in the general
case. It will be shown how the information from multiple cameras can
be used to constrain the remaining degrees of freedom. The
transition of our theory from the continuous into the discrete
domain results in effects that need to be accounted for during the
estimation process. Based on our theory, three independent
techniques for 3D motion estimation will be presented. The most
recently developed technique is based on a maximum likelihood
estimation with an integrated phase estimation method. A
hierarchical detection method will be proposed that increases the
computational effetiveness without compromising the estimation
accuracy. Experimental results will be presented and an outlook on
future research will be given.
Biography:Stefan Lehmann graduated from the Darmstadt University of Technology
with a German Diploma Degree in Electrical Engineering. During his
time in the United States, he worked as a Researcher in the
Department for Digital Security at the Fraunhofer Center for
Research in Computer Graphics in Providence, Rhode Island and as a
Guest Professor in the International Certificate Program for New
Media at the Rhode Island School of Design. Later on, he worked as
a Researcher in the Department for Virtual Reality and Scientific
Visualization at the Fraunhofer Institute for Computer Graphics in
Darmstadt, Germany and was involved in Lecturing activities at the
Darmstadt University of Technology. He is currently undertaking a
Ph.D. at the University of Queensland in the area of signal and
image processing. His interests in the signal and image processing
field are in 3D motion estimation and tracking techniques.
Type: Ph.D confirmation
Contact:Vaughan Clarkson, seminar host (vaughan@itee.uq.edu.au)
or Guido Governatori (ITEE seminar co-ordinator)
(guido@itee.uq.edu.au)