crisp slides v2
TRANSCRIPT
![Page 1: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/1.jpg)
CRISP: Coopera-on among Smartphones to Improve Indoor Posi-on Informa-on
Chen Qiu and Ma& W. Mutka Department of Computer Science and Engineering Michigan State University
IEEE WoWMoM 2015, Boston
![Page 2: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/2.jpg)
Everyone has a smartphone
![Page 3: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/3.jpg)
Outdoor Localiza-on Indoor Localiza-on
GPS
Ultrasound RFID
WSNs WiFi
What’s Next ?
![Page 4: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/4.jpg)
Dead Reckoning Approach
g
S1S2
SnSn−1
a x
a! a
!
g
O X
Y
Z
ay
az
a!= (ax ,ay ,az − g) Sn
!"!− Sn−1! "!!
= 12an−1! "!!
t 2 + vn−1! "!!
tnUniformly Accelerated Mo-on
![Page 5: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/5.jpg)
Drawback of Dead Reckoning
Dead Reckoning relies on the measurement accuracy of accelerometer
Accelerometer records the acceleration of the mobile device rather than a person’s body
![Page 6: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/6.jpg)
Drawback of Dead Reckoning Sensor used for localization, UM6 Small orientation errors cause serious deviation 0.5 degree error of the orientation sensor -> 308 meters error can occur within 1 minute
h&ps://www.pololu.com
![Page 7: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/7.jpg)
Challenge
For user of the smartphone, could we enhance the position accuracy for dead reckoning?
![Page 8: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/8.jpg)
• When a user encounters other users, by sharing locaDon and signal strength, all of their locaDon accuracies can be improved
(Location, Signal Strength)
Our Solu-on
LocalizaDon Error
Time
LocalizaDon Error
Time
(Location, Signal Strength) Alice Bob
![Page 9: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/9.jpg)
Received Signal Strength Indicator
Euclidean Distance
Our Solu-on • User’s iniDal posiDon is obtained from dead reckoning
• How to use the Received Signal Strength (RSS) ?
![Page 10: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/10.jpg)
Distance and RSSI Traditional formula is not accurate • various obstructions • multipath effect • other factors
Measure the distance and RSSI • Train for different devices • Store in Hashmap on a smartphone
Distance (meter) RSSI (dBm) 1m -‐40dBm 2m -‐45dBm
Distance (meter) RSSI (dBm)
1m -‐42dBm 2m -‐46dBm
Device 1 <-> Device 2 Device i <-> Device j
… …
d = 10[(P0−Fm−Pr−10×n×log10 ( f )+30×n−32.44)/10×n]
![Page 11: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/11.jpg)
• Locate a user’s position by knowing other devices’ locations and RSSI values
• After times of iterations, the deviation of each device will converge
• Encounter more other smartphones, the accuracy will be more accurate
Triangula-on Model
A Novel Perspective:
![Page 12: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/12.jpg)
RSSI= -48dBm
RSSI = -69dBm
RSSI = -75dBm
Dist(A,B)=4.5m
Dist(A,C)=5.5m
Dist(B,C)=2m
Mapping
Mapping
Mapping
(x , y )b b
(x , y )c c
(x , y )a a
(?, ?)
TriangleEquation
Carson
Bob
Alice
Triangula-on Model
![Page 13: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/13.jpg)
Triangula-on Calibra-on
1m
2m
Carson(C)
Bob (B) Alice (A)
$¶%¶
&¶
2m2m
$¶¶
ABC: the ground truth$¶%¶&¶: triangle with initial error$¶¶: estimated position bytriangle approach
![Page 14: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/14.jpg)
0 5 100
0.5
1
Time (seconds)
CD
F
ABC ErrorDead Reckoning
0 5 100
50
100
Time (seconds)Erro
r of D
ista
nce
(dec
imet
ers)
DeadReckoningAB ErrorBC ErrorAC ErrorABC Error
Preliminary Observa-ons
0 200 400 6000
1
2
Time (seconds)Erro
r of D
ista
nce
(m
eter
s)
Device ADevice BDevice C
0 100 200 300 400 5000
1
2
3
Time (Seconds)Er
ror o
f D
ista
nce
(met
ers)
• Within time increasing, the deviation is reduced effectively
![Page 15: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/15.jpg)
Geometry Extension • How about more than three users ? • How about Quadrilateral, Pentagon, Hexagon … ? • Decompose to triangle
Carson
Bob Alice(x , y )b b
(x ,y )c c
(x ,y )a a
David(x , y )d d
Carson
BobAlice
Bob Alice
David
Carson
Alice
David
![Page 16: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/16.jpg)
Feature of different signals • Bluetooth sensiDve to interference, also sensiDve to the distance low energy, supported by common smartphones, sampling frequency is not enough (>10s)
• WiFi supported by common smartphones sensiDve to interference, not sensiDve to the distance sampling frequency is not enough (>4s) • Zigbee sensiDve to interference, also sensiDve to the distance sampling frequency is enough not supported by common smartphones
![Page 17: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/17.jpg)
Combine different signals
WiFi + Bluetooth
Cloud�Server
Zigbee Device ID Time Stamp RSSI
Bluetooth RSSI-Distance Mapping
Distance
Device ID Time Stamp RSSI
Zigbee RSSI-Distance Mapping
Distance
Device ID Time Stamp RSSI
Bluetooth
Zigbee
WiFi Filter
Synchronous
Data Sample Format
![Page 18: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/18.jpg)
WiFi Direct Filter
0 50 100 150 200−90
−80
−70
−60
−50
−40
−30
−20
Time (seconds)
RSS
I (dB
m)
WiFiBluetoothBluetooth (Filtered)ZigbeeZigbee (Filtered)
NP NP
WiFi Direct enables devices to connect with each other without requiring a wireless access point
Replace the abnormal RSSI in the Noise Period
![Page 19: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/19.jpg)
Extra Benefit: Step Coun-ng
• Electronic pedometer on the smartphones
• Measuring steps by the accelerometer
![Page 20: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/20.jpg)
Extra Benefit: Step Coun-ng • Accelerometer is not reliable for pedometers
• Even if a user does not move, the received
acceleration can be changed sharply
![Page 21: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/21.jpg)
Extra Benefit: Step Coun-ng ! Users obtained the location continuously in different periods
! In each time period, we assume people walk straight
! Count steps for a user:
number of steps = ( moving distance / step length )
! Add the number of steps in each time period, the user can determine the number of steps they walked in total
100 200 300 400 5000
20
40
60
80
100
Time (seconds)
Erro
r Num
ber
of S
teps
ACCUPEDO
Noom Walk
CRISP
![Page 22: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/22.jpg)
System Evalua-on
Experiment Setup:
Data Collection in a Room Data Collection in a Hallway
! Train and collect data in Engineering Building of MSU ! More than 100 rooms and hallways, 6 volunteers ! Samsung S5, S3, Google Nexus, etc ! Android 4.4
![Page 23: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/23.jpg)
System Evalua-on
0 10 20 30 40 50
0
1
2
3
Time (per 20 seconds)
Erro
r of D
ista
nce
(
met
ers)
Triangle Approach
Combination Approach
Dead Reckoning Approach
5 10 15 20 25 30 35 40 450
2
4
6
8
10
Time (minutes)
Erro
r of D
ista
nce
(met
ers)
Dead ReckoningTriangle AppCombine App
Room measurement one time Room measurement 200 times
5 10 15 20 25 30 35 40 450
2
4
6
8
Time (minutes)
Erro
r of D
ista
nce
(met
ers)
Dead ReckoningTriangle AppCombine App
Hallway measurement one time
0 10 20 30 40 50
0
1
2
3
Time (per 20 seconds)
Erro
r of D
ista
nce
(m
eter
s)
Dead Reckoning Approach
Triangle Approach
Combination Approach
Hallway measurement 200 times
![Page 24: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/24.jpg)
System Evalua-on
Measurement in a complex environment
Dead Reckoning Trace Combination Approach Trace
����������
����������
100 200 300 400 5000
2
4
6
8
10
Time (seconds)
Erro
r of D
ista
nce
(m
eter
s)
Combination Approach
Dead Reckoning Approach
Triangle Approach
![Page 25: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/25.jpg)
10 20 30 40 500
1
2
3
4
Time (Minutes)
Erro
r of D
ista
nce
(met
ers)
BluetoothBluetoth+ZigbeeBluetooth+Zigbee+WiFi
0 10 20 30 40 50
0
0.5
1
Time (per 20 seconds)
Erro
r of D
ista
nce
(met
ers)
Bluetooth + Zigbee + WiFi
Bluetooth + Zigbee
Bluetooth
System Evalua-on
Comparison of different types of signals
0 100 200 300 400 500
0
0.5
1
1.5
Time (seconds)
Erro
r of D
ista
nce
(met
ers)
2+1 devices
3+1 devices
0 10 20 30 40 50−1
0
1
2
3
4
Time (per 20 seconds)
Erro
r of D
ista
nce
(met
ers)
2 assisted customers
no assisted customer
3−4 assisted customers
Comparison of the number of smartphones’ users
![Page 26: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/26.jpg)
Contribu-on Highlight
! Interact with other scanned smartphones to improve a user’s own localizaDon accuracy
! Improve the accuracy of the pedometer on a smartphone by RSSI rather than acceleraDon
! Combine the RSSI from Zigbee, WiFi and Bluetooth, and design a WiFi filter to reduce the noise
![Page 27: crisp slides v2](https://reader031.vdocuments.net/reader031/viewer/2022021813/5875a7ae1a28ab6d198b818f/html5/thumbnails/27.jpg)
Contact InformaDon: [email protected] eLANS Lab, CSE Dept. MSU