Visual Motion Analysis by Probabilistic Propagation of
Conditional Density
Michael Isard
D.Phil. Thesis, Oxford University, 1998.
Abstract
This thesis establishes a stochastic framework for tracking
curves in visual clutter, using a Bayesian random-sampling
algorithm. The approach is rooted in ideas from statistics,
control theory and computer vision. The problem is to track
outlines and features of foreground objects, modelled as curves,
as they move in
substantial clutter, and to do it at,
or close to, video frame-rate. The algorithm, named
Condensation, for
Conditional
density propag
ation, has
recently been derived independently by several researchers, and
is generating significant interest in the statistics and signal
processing communities. This thesis contributes to the
literature on
Condensation-like filters by
presenting some novel applications of and extensions to the
basic algorithm, and contributes to the visual motion estimation
literature by demonstrating high tracking performance in
cluttered environments. Despite its power the
Condensation algorithm has a remarkably simple
form and this allows the use of non-linear motion models which
combine characteristics of discrete Hidden Markov Models with
the continuous Auto-Regressive Process motion models
traditionally used in Kalman filters. These mixed
discrete-continuous models have promising applications to the
emerging field of perception of action. This thesis also
implements two algorithms to smooth the output of the
Condensation filter which improves the accuracy
of motion estimation in a batch-mode procedure after tracking is
complete.
Compressed
(gzip) postscript version of the thesis (15.4Mb).
The following pages will not reproduce correctly in black and
white:
64, 70-75, 80, 84, 86, 91, 92, 103, 106,
109-112
If you have the PSUtils package
you can select only the colour pages (12.5Mb) with the following command (note the page
numbers reflect the presence of the contents section):
psselect -p70,76-81,86,90,92,97,98,109,112,115-118
thesis.ps > colourpages.ps
and only the black and white pages (3Mb) as follows:
psselect
-p1-69,71-75,82-85,87-89,91,93-96,99-108,110,111,113,114,119-
thesis.ps > bwpages.ps
Alternatively, here is a version
(3.6Mb) with all of the pages, but with the large colour
images blanked out.
Back to
Michael Isard's home page