Crowdsourcing Based Indoor
Localization
Shen Ruofei
2015.3.10--present
Outline
S.J.T.U.
 Introduction
 Background
 Related Work
 Motivation
 Incomplete Database
 Future Work
Background
S.J.T.U.
Wi-Fi fingerprinting based indoor
localization
Location
2
Location
Fingerprint
1
Location
3
Fingerprints
collection
Fingerprint
Location
Fingerprint
4
Training Stage
Fingerprint
Fingerprints
database
Fingerprint
Localization Stage
Location
3
Fingerprints
database
Related Work
S.J.T.U.
Probabilistic Models Used for Indoor
Localization
Crowdsourcing based Indoor Localization
[1] K. Chintalapudi, A. P. Iyer, V. N. Padmanabhan, “Indoor localization without the pain”
, in Proc. ACM MobiCom, 2010, pp. 173-184.
[2] Z. Yang, C. Wu and Y. Liu, “Locating in fingerprint space: Wireless indoor
localization with little human intervention”, in Proc. ACM MobiCom, 2012, pp. 269–
280.
[3] C. Wu, Z. Yang, Y. Liu and W. Xi, “WILL: Wireless indoor localization without site
survey”, in Proc. INFOCOM, 2012, pp. 64-72.
Motivation
S.J.T.U.
Previous works are all done over a perfect
database, what the results would be if the
database is incomplete?
Outline
S.J.T.U.
 Introduction
 Incomplete Database
 System Model
 Error Probability
 Best Strategy
 Future Work
System Model
S.J.T.U.
Received Signal Strength
: How RSS readings varies with respect to locations
: Normalized Gaussian additive noise
System Model
S.J.T.U.
Sample Space & Physical Space
physical space:
Radius: δ
User’s location: Q
sample space:
Physical space
Event: E(δ)
sample space
System Model
S.J.T.U.
Data Fluctuation due to the Gaussian Noise
one dimension case
Law of
large
numbers
System Model
S.J.T.U.
Data Fluctuation due to the Gaussian Noise
Two dimension case
System Model
S.J.T.U.
Real database
n dimensional
Gaussian distribution
Error Probability
S.J.T.U.
Integration Area
Direct Integration Area:
①+②+③
Indirect Integration Area:
（①+④+③）+（②-④）
Error Probability
S.J.T.U.
Intermediate Integration Results
p
r2 y  r1 y
L
h
tan 
0
 2
de1 
 tan  e1
h
1
e
2
2

e12  e2 2
2 2
de2
Some variables
r1  (r1x , r1 y )
r2  (r2 x , r2 y )
tan   sin   
y0  x0 cot 
h
cot 
sin  
y0 
r2 y  r1 y
L
r1 y  r2 y
2
r1x  r2 x
x0 
2
Error Probability
S.J.T.U.
Error Probability
P




r2 y  r1 y
h
L


N



4 2
  
e1
2


1
2
2
e
e12  e22
2 2

e
4
N ( r12x  r12y  r22x  r22y )
2 2
hL
2 r2 y  r1 y de1
de2 dr1x dr1 y dr2 x dr2 y 
0
2
3
2
 LN
Error Probability
S.J.T.U.
Upper Bound and Lower Bound
y  cot  x0
y  cot  x0
1
erf ( 0
)erf ( 0
)  p2
4
2
cot  2
y0  cot  x0
y0  cot  x0
1
 erf (
)erf (
)
2
2
cot  2
Error Probability
S.J.T.U.
Simulation Results
Best Strategy
S.J.T.U.
After the establishment of database, we
should choose which AP RSS data are
superior than others

 n*  arg max (  Z i ) 2 
 n n
 i n

Zi 

i n

2
Best Strategy
S.J.T.U.
Greedy Algorithm
N
2 PN 1  Pi  2 PPN 1  cos(2 ) cos(2N 1 )  sin(2 )sin(2N 1 ) 
i 1
Dynamic Programming
Outline
S.J.T.U.
 Introduction
 Incomplete Database
 Future Work
 Experiment
 Best Strategy
Experiment
S.J.T.U.
Android Application experiment
Use the Android
Application to
establish a real
database to
verify our theory
work
Experiment
S.J.T.U.
The Marauder‘s Map(活点地图)
Best Strategy
S.J.T.U.
Whether best strategy is a NP question or
whether we could use the Dynamic
Programming method to solve this problem?
S.J.T.U.
Q&A
S.J.T.U.
Thank You