BraMBLe: A Bayesian MultipleBLob Tracker

1
BraMBLe A Bayesian Multiple-BLob Tracker

• M. Isard and J. MacCormick
• 2001

2
Results

• First particle filter tracking dynamic number of
  objects
• Runs real-time on extremely modest hardware
  (2001!)

3
We have
Object configurations
Observations
4
We have
Object configurations
Observations
We want
Posterior estimate
5
From Bayes Rule
6
From Bayes Rule
7
(No Transcript)
8
Fixed grid of points
9
Filter bank representation courtesy of Kristin
Branson
10
Response vector
Filter bank representation courtesy of Kristin
Branson
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
Brighter pixels are more likely to be in the
foreground
16
From Bayes Rule
17
Object configurations
18
(No Transcript)
19
(No Transcript)
20
(No Transcript)
21
(No Transcript)
22
From Bayes Rule
23
(No Transcript)
24
(No Transcript)
25
Image courtesy of Kevin Smith
26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
Track confusion
31
(No Transcript)
32
Summary

• Pros
• Tracking of variable numbers of objects
• Runs in real-time on old hardware
• Cons
• Requires a static camera
• Severe occlusions may cause track confusion

33
Questions?