Introduction To Particle Filters: A Trajectory Tracking Example

Definitions and issues
Particle filtering algorithms

Real applications
Introduction to particle filters:

a trajectory tracking example
Alexis Huet
23 May 2014
1/31
Alexis Huet
2/31

Real applications
Outline

Markov chains
Hidden Markov models
Issues

Sequential Importance Sampling algorithm
Sequential Importance Sampling Resampling algorithm
Comparison of algorithms
Real applications
Alexis Huet
3/31

Real applications
Markov chains
Issues
Outline

Markov chains
Issues

Real applications
Alexis Huet

Real applications
Markov chains
Issues
Markov chains
We consider the following state space : (Rn , L(Rn )).
Definition
A sequence of random variables (Xk )kN taking values in Rn is a
Markov chain if for all k 1, x0 , . . . xk1 Rn and A L(Rn ) :
P(Xk A|Xk1 = xk1 , . . . , X0 = x0 ) = P(Xk A|Xk1 = xk1 ).
Hypothesis
In the sequel, the Markov chain is homogeneous and admits a
family of density functions (p(.|x))xRn such that :
Z
P(Xk A|Xk1 = xk1 ) =
p(xk |xk1 )dxk .
A
4/31
Alexis Huet

Real applications
Markov chains
Issues
Example
The state space is here R2 . For all x R2 , let B(x, 1) be the unit
ball centered in x.
Definition
For all x R2 , the distribution Unif (B(x, 1)) is defined by its
density :
1
1(. B(x, 1)).
For the initial condition, we take :

X0
Unif (B(0, 1)).
And for transition distributions :

Xk |(Xk1 = xk1 )
5/31
Alexis Huet
Unif (B(xk1 , 1)).

Real applications
Markov chains
Issues
0
2
ord
Example
abs
6/31
Alexis Huet

Real applications
Markov chains
Issues
Example
0
2
ord
abs
6/31
Alexis Huet

Real applications
Markov chains
Issues
Example
0
2
ord
abs
6/31
Alexis Huet

Real applications
Markov chains
Issues
Example
ord
abs
6/31
Alexis Huet

Real applications
Markov chains
Issues
Example
ord
abs
6/31
Alexis Huet

Real applications
Markov chains
Issues
Example
ord
abs
6/31
Alexis Huet

Real applications
Markov chains
Issues
Example
03
ord
6
1
abs
6/31
Alexis Huet

Real applications
Markov chains
Issues
Definition
(Xk , Yk )k0:m1 is a hidden Markov model if : (Xk )k0:m1 is a
Markov chain, (Yk )k0:m1 are independent conditionally to
(Xk )k0:m1 and for all k, Yk depends only on Xk .
Schematically, we have :
7/31
X0
X1
X2
. . . Xm1
y
y .
Y0
Y1
Y2
...
Alexis Huet
Ym1

Real applications
Markov chains
Issues
Example
We choose (Yk ) taking values in ] , ]. Conditionally to

Xk = xk , we define :
Yk = Arg (xk ) +
where
N (0, 2 ).
Related density : yk 7 p(yk |xk ).
8/31
Alexis Huet

Real applications
Markov chains
Issues
Example ( = 0.1)
03
03
2 4
ord
5 1
6
8
abs
9/31
Alexis Huet

Real applications
Markov chains
Issues
Example ( = 0.1)
03
ord
6
9
abs
10/31
Alexis Huet

Real applications
Markov chains
Issues
Issues
Now, the hidden chain (xk ) is unknown and we only have the
observations (yk ).
Aim : reconstitute x0:m1 := (x0 , . . . , xm1 ) conditionally to the
observations.
Specifically, to generate a sample according to the following
density :
p(x0:m1 |y0:m1 )dx0:m1 .
11/31
Alexis Huet

Real applications

Outline
12/31

Markov chains
Issues

Real applications
Alexis Huet

Real applications

Sequential Importance Sampling (SIS) algorithm
We try to simulate a sample from the density p(x0 |y0 )dx0 .

p(x0 |y0 ) =
1
p(x0 , y0 )
=
p(y0 |x0 )p(x0 ).
p(y0 )
p(y0 )
We can generate a sample from p(x0 )dx0 .

We can compute x0 7 p(y0 |x0 ).
13/31
Alexis Huet

Real applications
p(x0 |y0 )dx0 =

1
p(y0 |x0 )p(x0 )dx0 .
p(y0 )
SIS algorithm (first step) :

Generate a sample of length N from p(x0 )dx0 :
(1)
(N)
x0 , . . . , x0 .
Compute for each particle j :
(j)
w0 =
1
(j)
p(y0 |x0 ).
p(y0 )
(j)
The sample (
x0 )j1:N which approximates p(x0 |y0 )dx0 is :
N
X
j=1
14/31
(j)
w0
PN
(j)
(j 0 ) x0
j 0 =1 w0
Alexis Huet
(dx0 ).

Real applications
p(x0 |y0 )dx0 =

1
p(y0 |x0 )p(x0 )dx0 .
p(y0 )
SIS algorithm (first step) :

Generate a sample of length N from p(x0 )dx0 :
(1)
(N)
x0 , . . . , x0 .
(j)
(j)
w0 = p(y0 |x0 ).
(j)
The sample (
x0 )j1:N which approximates p(x0 |y0 )dx0 is :
N
X
j=1
15/31
(j)
w0
(j)
(j 0 ) x0
j 0 =1 w0
PN
Alexis Huet
(dx0 ).

Real applications

Example (SIS) N = 20
0.070.93
0.00
0.00
0.00
0.00
ord
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.000.00
0.00
0.00
0.00
0.00
0.00
1.0
0.5
0.0
0.5
1.0
1.5
abs
16/31
Alexis Huet
2.0

Real applications

p(x0:i |y0:i )dx0:i =

1
p(y0 |x0 ) . . . p(yi |xi )p(x0 )p(x1 |x0 ) . . . p(xi |xi1 )dx0:i .
p(y0:i )
SIS algorithm (step i). For all particle j 1 : N :
(j)
Conditionally to xi1 , generate a sample according to
p(xi |xi1 )dxi :
(1)
(N)
xi , . . . , xi .
(j)
wi
(j)
(j)
= wi1 p(yi |xi ).
(j)
The sample (
x1:i )j1:N which approximates p(x1:i |yi )dx1:i is :
N
X
j=1
17/31
(j)
wi
PN
(j)
(j 0 ) x1:i
j 0 =1 wi
Alexis Huet
(dx1:i ).

Real applications

0.070.93
0.00
0.00
0.00
0.00
ord
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.000.00
0.00
0.00
0.00
0.00
0.00
1.0
0.5
0.0
0.5
1.0
1.5
abs
18/31
Alexis Huet
2.0

Real applications

0.00
0.00
0.00
0.00
0.00
ord
0.00
0.00
0.00
0.00
0.00
0.00
1.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
1.0
0.5
0.0
0.5
1.0
1.5
abs
18/31
Alexis Huet
2.0

Real applications

1.00
0.00
0.00
0.00
ord
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
1.0
0.5
0.0
0.5
1.0
1.5
abs
18/31
Alexis Huet
2.0

Real applications

Convergence and degeneracy problem (SIS)
Convergence :
for m the length of the chain fixed, central limit theorem
when the number of particles N +.
Degeneracy problem :
many particles have a relative weight close to 0%,
for N fixed, from a certain number of sites, only one particle
has an important relative weight.
Resampling of the particles.
19/31
Alexis Huet

Real applications
p(x0 |y0 )dx0 =

1
p(y0 |x0 )p(x0 )dx0 .
p(y0 )
SISR algorithm (first step)

Generate a sample of length N according to p(x0 )dx0 :
(1)
(N)
x0 , . . . , x0 .
(j)
(j)
w0 = p(y0 |x0 ).
(j)
Generate (
x0 )j1:N from :
N
X
j=1
(j)
w0
(j)
and let (w
0 )j1:N 1.
20/31
(j)
(j 0 ) x0
j 0 =1 w0
PN
Alexis Huet
(dx0 )

Real applications

SISR algorithm (step i). For all particle j 1 : N :

(j)
Conditionally to xi1 , generate a sample according to

p(xi |
xi1 )dxi :
(1)
(N)
xi , . . . , xi .
(j)
wi
(j)
(j)
(j)
=w
i1 p(yi |xi ) = p(yi |xi ).
(j)
Generate (
xi )j1:N from :
N
X
j=1
(j)
wi
(j)
and let (w
i )j1:N 1.
21/31
(j)
(j 0 ) xi
j 0 =1 wi
PN
Alexis Huet
(dxi )

Real applications

Example (SISR)
1.0
0.00
0.00
0.00
0.00
0.00
0.00
0.72
0.00 0.28
0.00
0.00
0.00
0.0
ord
0.5
0.00
0.00
0.00
0.00
0.5
0.00
0.00
0.00
0.00
1.0
1.0
0.5
0.0
0.5
abs
22/31
Alexis Huet
1.0

Real applications

1.0
Example (SISR)
0.0
1.0
0.5
ord
0.5
0.05
0.05
1.0
0.5
0.0
0.5
abs
22/31
Alexis Huet
1.0

Real applications

Example (SISR)
0.00
1.0
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.5
0.00
0.00
0.04
0.03
0.22
0.36
0.35
0.0
ord
0.00
0.00
0.00
1.0
0.5
1.0
0.5
0.0
0.5
abs
22/31
Alexis Huet
1.0

Real applications

1.0
Example (SISR)
0.5
0.0
1.0
0.5
ord
0.05
0.05
0.05
0.05
1.0
0.5
0.0
0.5
abs
22/31
Alexis Huet
1.0

Real applications

Convergence (SISR)
Elimination of the weights degeneracy problem.

Convergence :
for m the length of the chain fixed, central limit theorem
when the number of particles N +.
23/31
Alexis Huet

Real applications

Example
Comparison with m = 100 sites and N = 10000 particles.
84 85
4
90
80
97
86
82
96
9883
81
87
99
95
94
88
92
93
89
91
79
78
77
74
75
73
76
72
03
4
71
5 1
67
ord
270
68
69
66
61
65
10
6
7
12 63
11
1713
18 62
8
19
60 9
14
64

59 36 38
20
2829 33
22
16
15
32
35
23
58 34
24
37
39
48
47
21
40
26 3027 49 31
25
43
57
56
46
44
55
45
50
54
53
51
52
abs
24/31
Alexis Huet
41
42

Real applications
Example (SIS)
24/31
Alexis Huet


Real applications
Example (SISR)
24/31
Alexis Huet


Real applications
Example (SIS)
25/31
Alexis Huet


Real applications
Example (SISR)
25/31
Alexis Huet


Real applications
Outline
26/31

Markov chains
Issues

Real applications
Alexis Huet

Real applications
Path tracking
Observations : noisy position.

Hidden states : real position.
Parameters : behavior of the moving body.
27/31
Alexis Huet

Real applications
Voice recognition system
Observations : a word is pronounced, cut every 15ms.

Hidden states : phonemes that led to this pronounced word.
Parameters : the set of all dictionary words.
28/31
Alexis Huet

Real applications
Phylogenetic analysis
Observations : DNA sequences of several species at the leafs

of a tree graph.
Hidden states : all DNA sequences from the common ancestry
sequence to the present time.
Parameters : mutation parameters, lengths of the tree
branches.
29/31
Alexis Huet

Real applications
30/31
t = 1
ACGAGGTGA
ACAAGGTGA
ACAGGGTGA
ACAGGGCGA
ACAGGGCAA
t=0
ACAGGGCAA
Alexis Huet

Real applications
30/31
i
t = 1
1
A
A
A
A
A
2
C
C
C
C
C
3
G
A
A
A
A
4
A
A
G
G
G
5
G
G
G
G
G
6
G
G
G
G
G
7
T
T
T
C
C
8
G
G
G
G
A
9
A
A
A
A
A
t=0
Alexis Huet

Real applications
30/31
i
t = 1
1
?
?
?
?
?
2
?
?
?
?
?
3
?
?
?
?
?
4
?
?
?
?
?
5
?
?
?
?
?
6
?
?
?
?
?
7
?
?
?
?
?
8
?
?
?
?
?
9
?
?
?
?
?
t=0
Alexis Huet

Real applications
Olivier Cappe, Eric Moulines, and Tobias Ryden. Inference in

hidden Markov models. Springer, 2005.
Neil Gordon, David Salmond, and Adrian Smith. Novel
approach to nonlinear/non-Gaussian Bayesian state
estimation. IEEE Proceedings F (Radar and Signal
Processing), 1993.
Lawrence Rabiner. A tutorial on hidden Markov models and
selected applications in speech recognition. Proceedings of the
IEEE, 1989.
31/31
Alexis Huet

Introduction To Particle Filters: A Trajectory Tracking Example

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction To Particle Filters: A Trajectory Tracking Example

Uploaded by

Copyright:

Available Formats

Definitions and issues

Particle filtering algorithms