PreliminariesBackground Subtraction GregMori CMPT888 Outline - PowerPoint PPT Presentation

Preliminaries
–
Background
 Subtraction
 Greg
Mori
 CMPT888


Outline
 • A
little
bit
on
background
subtraction


Background
Subtraction


Basic
Pipeline
 Background
 Mixture
of
 Background
 Gaussians
 Image
 Model
 Median
value
 EM
fitting
 Likelihood
under
 Absolute
value
of
 difference
with
 of
each
pixel
 background
model,
 over
time
 background,
thresholded
 thresholded
(or
not)


Many
Pitfalls
 • Background
subtraction
is
not
perfect
in
 practice
 – Background
changes
 – Foreground
looks
like
background
 – Objects
stop
moving
 – …
 • But
it’s
a
useful
signal,
combine
into
human
 detector


Summary
 • For
more
information:
 – Chris
Stauffer
and
W.E.L.
Grimson.
Adaptive
background
 mixture
models
for
real‐time
tracking.
CVPR
1999
 – Kentaro
Toyama,
John
Krumm,
Barry
Brumitt,
and
Brian
 Meyers.
Wallflower:
Principles
and
practice
of
background
 maintenance.
ICCV
1999