Two particle filters for multiple target tracking that make use of auxiliary filtering are introduced. The derivation of the particle filters is addresed in the paper "Generalizations of the auxiliary particle filter for multiple target tracking" that was accepted for oral presentation in the 17th International Conference on information Fusion, "Fusion 2014" in Salamanca.
Generalizations of the auxiliary particle filter for multiple target tracking
1. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Generalizations of the auxiliary particle filter for
multiple target tracking
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal
Dpto. Se˜nales, Sistemas y Radiocomunicaciones, Universidad Polit´ecnica de Madrid,
Spain
†Dept. of Electrical and Computer Engineering, Curtin University, Australia
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
2. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Outline
1 Particle filters for Multiple target tracking
2 State partition particle filters
3 Generalizations of auxiliary particle filtering for MTT
APP
TRAPP
4 Simulations and results
Target dynamics and sensor modeling
Results
5 Conclusions
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
3. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Outline
1 Particle filters for Multiple target tracking
2 State partition particle filters
3 Generalizations of auxiliary particle filtering for MTT
APP
TRAPP
4 Simulations and results
Target dynamics and sensor modeling
Results
5 Conclusions
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
4. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Multiple target tracking.
Multiple target tracking
MTT is usually formulated in the Bayesian framework. The
information of interest about the targets is contained in the
multitarget posterior PDF.
Multitarget state
Xk
= (xk
1)T
, (xk
2)T
, ..., (xk
t )T
T
∈ Rn·t
Posterior PDF
p(Xk
|z1:k
)
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
5. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Particle filters.
PFs sample the state-space to build an approximation to the
posterior PDF.
The dimension of the state-space linearly grows with the
number of targets.
Sampling high-dimension state-spaces is very inefficient, giving
rise to the curse of dimensionality.
Some modifications are needed if PFs are to be successfully
applied to MTT.
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
6. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Outline
1 Particle filters for Multiple target tracking
2 State partition particle filters
3 Generalizations of auxiliary particle filtering for MTT
APP
TRAPP
4 Simulations and results
Target dynamics and sensor modeling
Results
5 Conclusions
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
7. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
State partition.
State partition
To appease the curse of dimensionality some algorithms assume posterior
independence between targets. This allows for the partition of the state-space to
individually sample the state of each target.
p(X
k+1
|z
1:k+1
) =
t
j=1
pj (x
k+1
j |z
1:k+1
)
Some algorithms that work under the independence assumption are:
Independent Joint Optimal Importance Density PF (IJOID) [1].
Independent Partition PF (IP) [2].
Parallel Partition PF (PP) [3].
[1] W. Yi, M. R. Morelande, L. Kong, and J. Yang, “A computationally efficient particle filter for multitarget
tracking using an independence approximation,” IEEE Transactions on Signal Processing, Feb. 2013.
[2] M. Orton and W. Fitzgerald, “A Bayesian approach to tracking multiple targets using sensor arrays and particle
filters,” IEEE Transactions on Signal Processing, 2002.
[3] A. F. Garc´ıa-Fern´andez, M. Morelande, and J. Grajal, “Two-layer particle filter for multiple target detection and
tracking,” IEEE Transactions on Aerospace and Electronic Systems, 2013.
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
8. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
Outline
1 Particle filters for Multiple target tracking
2 State partition particle filters
3 Generalizations of auxiliary particle filtering for MTT
APP
TRAPP
4 Simulations and results
Target dynamics and sensor modeling
Results
5 Conclusions
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
9. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
Auxiliary PFs for MTT.
When there is only one target present, both IP and PP come
down to the sequential-importance-resampling PF, which is
usually outperformed by the auxiliary particle filter.
Two particle filters, APP and TRAPP, are presented inspired
by the APF and the state-partition strategy of PP, resulting
on generalizations of the APF for multiple target tracking.
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
10. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
Auxiliary parallel partition.
APP
APP makes use of auxiliary particle filtering for each target,
selecting those subparticles at time k that are prone to generate
subparticles with higher target-likelihood at time k + 1 according
to zk+1 .
q(Xk+1
, a|z1:k+1
) =
t
j=1
qj (xk+1
j , aj |z1:k+1
)
qj (xk+1
j , aj |z1:k+1
) ∝ bj (µk+1
j,aj
)ωk
aj
p(xk+1
j |xk
j,aj
)
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
11. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
Auxiliary parallel partition. Example
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
12. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
Auxiliary parallel partition. Example
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
13. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
Auxiliary parallel partition. Example
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
14. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
Auxiliary parallel partition. Example
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
15. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
Auxiliary parallel partition. Example
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
16. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
Auxiliary parallel partition. Example
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
17. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
Auxiliary parallel partition. Example
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
18. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
Auxiliary parallel partition. Example
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
19. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
Target-resampling auxiliary parallel partition.
TRAPP PF
Target-resampling (as in IP and PP) is not always undesirable,
depending on the sensor model and the dimension of the state
space. TRAPP makes use of auxiliary filtering followed by
target-resampling.
q(Xk+1
, a|z1:k+1
) =
t
j=1
qj (xk+1
j , aj |z1:k+1
)
qj (xk+1
j , aj |z1:k+1
) ∝ bj (xk+1
j )ωk
aj
p(xk+1
j |xk
j,aj
)
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
20. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
TRAPP. Example
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
21. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
TRAPP. Example
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
22. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
TRAPP. Example
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
23. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
APP
TRAPP
IP, PP, APP and TRAPP
auxiliary
filtering in
target sampling
target resampling accounts
for nearby
targets
avoids
particle
resampling
IP × × ×
PP × ×
APP ×
TRAPP
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
24. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Target dynamics and sensor modeling
Results
Outline
1 Particle filters for Multiple target tracking
2 State partition particle filters
3 Generalizations of auxiliary particle filtering for MTT
APP
TRAPP
4 Simulations and results
Target dynamics and sensor modeling
Results
5 Conclusions
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
25. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Target dynamics and sensor modeling
Results
Target dynamics.
The trajectories of the targets are generated according to an
independent nearly-constant velocity model.
0 20 40 60 80 100 120
0
20
40
60
80
100
120
1
2
3
4
5
6
7
8
x position [m]
yposition[m]
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
26. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Target dynamics and sensor modeling
Results
Sensor model.
A nonlinear measurement model is considered.
zk+1
i = hi (Xk+1
) + vk+1
i
hi (Xk+1
) =
t
j=1
SNR(dk+1
j,i )
SNR(dk+1
j,i ) =
SNR0 dk+1
j,i ≤ d0
SNR0
d2
0
(dk+1
j,i )2
dk+1
j,i > d0
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
27. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Target dynamics and sensor modeling
Results
Compared filters
PP
APP
TRAPP
Jointly Auxiliary PF (JA)
Adaptive Auxiliary PF (AA)
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
28. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Target dynamics and sensor modeling
Results
Tracking 1 target.
50 100 150 200 250 300 350 400 450 500
0
0.5
1
1.5
2
2.5
3
3.5
Number of particles
RMSOSPApositionerror[m]
TRAPP
APP
PP
AA
JA
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
29. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Target dynamics and sensor modeling
Results
Tracking 8 targets.
50 100 150 200 250 300 350 400 450 500
0
2
4
6
8
10
12
14
16
Number of particles
RMSOSPApositionerror[m]
TRAPP
APP
PP
AA
JA
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
30. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Target dynamics and sensor modeling
Results
Tracking 1 to 8 targets, 100 particles.
1 2 3 4 5 6 7 8
0
2
4
6
8
10
12
14
16
Number of targets
RMSOSPApositionerror[m]
TRAPP
APP
PP
AA
JA
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
31. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Target dynamics and sensor modeling
Results
Tracking 1 to 8 targets, 100 particles. Narrow likelihood.
1 2 3 4 5 6 7 8
0
2
4
6
8
10
12
Number of targets
RMSOSPApositionerror[m]
TRAPP
APP
PP
AA
JA
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
32. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Outline
1 Particle filters for Multiple target tracking
2 State partition particle filters
3 Generalizations of auxiliary particle filtering for MTT
APP
TRAPP
4 Simulations and results
Target dynamics and sensor modeling
Results
5 Conclusions
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
33. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Conclusions
Two particle filters, APP and TRAPP, have been developed
that generalize the auxiliary particle filtering for multiple
target tracking, making use of the state-partition strategy
based on posterior independence.
Both APP and TRAPP outperform similar filters for MTT and
are generally applicable algorithms.
APP generally outperforms TRAPP, however, TRAPP can
outperform APP when dealing with some measurement and
dynamic models and a high number of targets.
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr
34. Particle filters for Multiple target tracking
State partition particle filters
Generalizations of auxiliary particle filtering for MTT
Simulations and results
Conclusions
Thank you
Luis ´Ubeda-Medina , ´Angel F. Garc´ıa-Fern´andez†
, Jes´us Grajal Generalizations of the auxiliary particle filter for multiple target tr