[CONTEXTS'11] A bayesian strategy to enhance the performance of indoor localization systems
0 likes319 views
The document presents a Bayesian strategy to enhance the performance of indoor localization systems. It describes applying a dynamic Bayesian network using proximity and transition sensors to estimate a user's location. A simulation of the strategy in a 6 zone deployment showed around a 16% improvement in location estimation accuracy over the existing location system alone. Future work could involve considering more sensor types and placements and performing a real system implementation.
![application scenario
x area id 4 6
WSN anchor node 1 RFID tag 3 RFID tag
1 proximity mote 1 proximity mote
transition sensors 2
1 RFID tray 1 RFID tray
t46
t34
t43
0 1 3 5
user t35
+ mobile mote
1 RFID tag 1 proximity mote
+ PDA + RFIDreader
1 RFID tray
existing location system objective: Bayesian fusion strategy
•symbolic location (zone-based)• output: 2)adding new proximity detection
•NZ = 6 zones (~ rooms) o(t)=0,1,…, NZ-1 sensors
•WSN network (ZigBee) • average error = 28.79% 3)adding new transition sensors
• accuracy model: between zones
•12 anchor nodes (2 x zone)
P(o(t)|Hk(t)) 4)information about the particular
[Hk(t) real user location] deployment (possible transitions)
User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 5 / 14](https://image.slidesharecdn.com/contexts11abayesianstrategytoenhancetheperformanceofindoorlocalizationsystems-120611064501-phpapp01/85/CONTEXTS-11-A-bayesian-strategy-to-enhance-the-performance-of-indoor-localization-systems-5-320.jpg)
![sensor models
• passive RFID
proximity
P(dn(t)|cn(t))
• pressure mats
sensors • power-tuned ZigBee motes [dn(t) proximity sensor state]
• etc. [cn(t) 1 if user in sensor proximity]
transition
• pair of pressure mats P(in(t)|rpq(t))
• power-tuned ZigBee motes [in(t) transition sensor state]
sensors
• etc.
[rpq(t) 1 if user transition exists]
User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 6 / 14](https://image.slidesharecdn.com/contexts11abayesianstrategytoenhancetheperformanceofindoorlocalizationsystems-120611064501-phpapp01/85/CONTEXTS-11-A-bayesian-strategy-to-enhance-the-performance-of-indoor-localization-systems-6-320.jpg)
![" 7(" ' -" ' / . # ' # " ' (&#&# M " *-A' # (" , -. , 78) "/ &, 6 +. 2 . =
# " +) % " $, &' # &# #
Bayesian strategy
! ! ! !! ! ! ! ! ! ! ! ! ! ! !!!!! ! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! !
!! !!
" ' -" ' / . # &% " ' (" # " =
1-' 2 8) #
transition model
!"!#$ %! ! ! ! !
&
# modelado de la calidad de los sensores modelo de transición obtenido al calcular
*(excepto para los de transición) [between zones]
subyacente
#
#
! ! ! ! ! ! ! "#!! " # ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! #
!
!!
#
#
!
!" #$ %! ! ! ! !
&
modelado de la calidad de los sensores modelo de transición obtenido al calcular
# sensors model recursive component
(slightly different for transition sensors; see paper)subyacente
*(excepto para los de transición)
[obtained when calculating Hk(t-1)]
1" +(. *# " # "*) % . # &% " %/ " ' . 2 -' &/ . , # " # *(&# +) &+-A' # B, &7 $) " *# *# 2 Z' # &# &% -" , # .
/ , (&/ 1-' 7 ## / " " *. # " +. $&, +) 8) E&%
"# ' -' ) " % *" "
/ -+" # C# . # 1% C" # ' #&# %++-A' # " # >M . 4
/ 2 -2 #
# 2 B&,0. 7$) " / " # , # # &' (" ' " , % # &# -/ &, # " #&# (-/ &/ # " # # *(># &M -3&' / . # ' # &#
" # *" Z(-%2 . $&, E&% 8) % +&' 8) *" " 2 -2 " +&/ 2
9 / 14
&# . B&B-% &/ # Technologies and Applications$%# " # *(># ' (, " #. *# . , " *# C# 4
$, User-Centric 8) " # 1" +(-E&2 " ' (" # 2 – " 8) "
-/ C# " +) CONTEXTS
" abernardos@grpss.ssr.upm.es
% E&% a# L #](https://image.slidesharecdn.com/contexts11abayesianstrategytoenhancetheperformanceofindoorlocalizationsystems-120611064501-phpapp01/85/CONTEXTS-11-A-bayesian-strategy-to-enhance-the-performance-of-indoor-localization-systems-9-320.jpg)
[CONTEXTS'11] A bayesian strategy to enhance the performance of indoor localization systems
- 1. Grupo de Procesado de Datos y Simulación ETSI de Telecomunicación Universidad Politécnica de Madrid A Bayesian strategy to enhance the performance of indoor localization systems CONTEXTS 2011 Josué Iglesias, Ana M. Bernardos, José R. Casar abernardos@grpss.ssr.upm.es
- 2. contents introduction application scenario (sensor models) Bayesian enhancement strategy simulation results discussion and future works User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 2 / 14
- 3. introduction • smart environments (AmI, context-awared, etc.) different heterogeneous technologies: – WSN – RFID data fusion techniques – bi-dimensional codes – etc. location estimation enhancement • smart environments indoor location services (based on infrared, ultrasounds, video, radio frequency, etc.) User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 3 / 14
- 4. contents ✓ introduction application scenario (sensor models) Bayesian enhancement strategy simulation results discussion and future works User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 4 / 14
- 5. application scenario x area id 4 6 WSN anchor node 1 RFID tag 3 RFID tag 1 proximity mote 1 proximity mote transition sensors 2 1 RFID tray 1 RFID tray t46 t34 t43 0 1 3 5 user t35 + mobile mote 1 RFID tag 1 proximity mote + PDA + RFIDreader 1 RFID tray existing location system objective: Bayesian fusion strategy •symbolic location (zone-based)• output: 2)adding new proximity detection •NZ = 6 zones (~ rooms) o(t)=0,1,…, NZ-1 sensors •WSN network (ZigBee) • average error = 28.79% 3)adding new transition sensors • accuracy model: between zones •12 anchor nodes (2 x zone) P(o(t)|Hk(t)) 4)information about the particular [Hk(t) real user location] deployment (possible transitions) User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 5 / 14
- 6. sensor models • passive RFID proximity P(dn(t)|cn(t)) • pressure mats sensors • power-tuned ZigBee motes [dn(t) proximity sensor state] • etc. [cn(t) 1 if user in sensor proximity] transition • pair of pressure mats P(in(t)|rpq(t)) • power-tuned ZigBee motes [in(t) transition sensor state] sensors • etc. [rpq(t) 1 if user transition exists] User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 6 / 14
- 7. contents ✓ introduction ✓ application scenario (sensor models) Bayesian enhancement strategy simulation results discussion and future works User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 7 / 14
- 8. Bayesian strategy Dinamic Bayesian Network real user location hidden states sensor observations transition location proximity sensors state system state sensors state recursive Bayesian filter • temporal hidden states transitions Markovian evolution • sensor observation independent (according the DBN graph) User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 8 / 14
- 9. " 7(" ' -" ' / . # ' # " ' (&#&# M " *-A' # (" , -. , 78) "/ &, 6 +. 2 . = # " +) % " $, &' # &# # Bayesian strategy ! ! ! !! ! ! ! ! ! ! ! ! ! ! !!!!! ! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! !! !! " ' -" ' / . # &% " ' (" # " = 1-' 2 8) # transition model !"!#$ %! ! ! ! ! & # modelado de la calidad de los sensores modelo de transición obtenido al calcular *(excepto para los de transición) [between zones] subyacente # # ! ! ! ! ! ! ! "#!! " # ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! # ! !! # # ! !" #$ %! ! ! ! ! & modelado de la calidad de los sensores modelo de transición obtenido al calcular # sensors model recursive component (slightly different for transition sensors; see paper)subyacente *(excepto para los de transición) [obtained when calculating Hk(t-1)] 1" +(. *# " # "*) % . # &% " %/ " ' . 2 -' &/ . , # " # *(&# +) &+-A' # B, &7 $) " *# *# 2 Z' # &# &% -" , # . / , (&/ 1-' 7 ## / " " *. # " +. $&, +) 8) E&% "# ' -' ) " % *" " / -+" # C# . # 1% C" # ' #&# %++-A' # " # >M . 4 / 2 -2 # # 2 B&,0. 7$) " / " # , # # &' (" ' " , % # &# -/ &, # " #&# (-/ &/ # " # # *(># &M -3&' / . # ' # &# " # *" Z(-%2 . $&, E&% 8) % +&' 8) *" " 2 -2 " +&/ 2 9 / 14 &# . B&B-% &/ # Technologies and Applications$%# " # *(># ' (, " #. *# . , " *# C# 4 $, User-Centric 8) " # 1" +(-E&2 " ' (" # 2 – " 8) " -/ C# " +) CONTEXTS " abernardos@grpss.ssr.upm.es % E&% a# L #
- 10. contents ✓ introduction ✓ application scenario (sensor models) ✓ Bayesian enhancement strategy simulation results discussion and future works User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 10 / 14
- 11. simulation results 5.2 Evaluation of the Bayesian location improvement algorithm in a real scenario simulating a future Figure 5 the results of a simulation employing togethermodel proximity and Finally, in real deployment: • proximity sensor both transition sensors are presented. This test has been set tocalculated for mote-based sensors): (empirically be run using our real •6 zone deployment: – deployment configuration (Figure 1). The number of–proximity(t)=0) = 1 has been set to 11 proximity sensors P(d (t)=0|c sensors n n – 4 transition sensors simulation (matching the number of sensors (t)=1|c (t)=1) = 0.978926 11 for this – P(d nowadays available for our n n •transition model: deployment). Only 4 transition sensors • have been sensor model placed in the real transition employed, – equidistributed if zones communicate – ranging from 85% to 100% of hit rate •location system quality decided to place there several transitionsimulation scenario: reduce location so it was • sensors, trying to (empirically calculated) = 71.21% hit rate still working in the configuration of the transition sensors, system errors. As we are – 1.000 trajectories – 1.000 zone transition per trajectory the obtained improvement is shown over several transition sensors qualities. hit rate (%): location system + transition model + 11 proximity sensors + 4 transition sensors location system + transition model ~ + 16 % + 11 proximity sensors location system + transition model + 4 transition sensors location system + transition model location system Fig. 5. Real deployment influence over location estimation User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 11 / 14
- 12. contents ✓ introduction ✓ application scenario (sensor models) ✓ Bayesian enhancement strategy ✓ simulation results discussion and future works User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 12 / 14
- 13. discussion & future works • AmI environments make use of several heterogeneous technologies (e.g., RFID, bi-dimensional codes, etc.) that can be seamless processed to enhance already deployed location systems – cheap and feasible approach – hit rate 71.21% ~ 88% • consider more types of sensors (e.g., RFID, pressure mats, etc.), empirically obtaining its probabilistic models • perform more tests with different sensor’s placements (analysing the enhancement introduced by each kind of sensor) • real implementation – design supporting infrastructure – mobile deployment? User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 13 / 14
- 14. any question? User-Centric Technologies and Applications – CONTEXTS abernardos@grpss.ssr.upm.es 14 / 14