Upload
roberta-anthony
View
221
Download
0
Tags:
Embed Size (px)
Citation preview
DAISYData Analysis and Information SecuritY Lab
E-eyes: Device-free Location-oriented Activity Identification Using Fine-grained
WiFi Signatures
Presenter: Yan Wang
Yan Wang†, Jian Liu†, Yingying Chen†, Marco Gruteser‡, Jie Yang#, Hongbo Liu*
†Dept. of ECE, Stevens Institute of Technology‡ WINLAB, Rutgers University
# Dept. of CS, Florida State University* Indiana University-Purdue University Indianapolis
MobiCom 2014Maui, Hawaii
Sept. 7th – 11th 2014
Motivation and Applications
2
Our Goal: Low-Cost Fine-Grained Activity Identification
3
Scalability / Infrastructural costLow cost High cost
Fin
e-g
rain
edC
oar
se-g
rain
ed
Gra
nu
lari
ty o
f th
e so
luti
on
s
Activity sensorsAttached sensors
Non-attached sensors
Localization/classificationusing specialized devices
WiSee, WiTrack
Localization
RTI
Localization using off-the-shelf WiFi
Device-free passive localization
RSS-based approach
Optimal solution
Our E-eyes
Nonintrusive Intrusive
Intuition and Basic Idea
4
Increasing availability of WiFi signals in home environments WiFi provides fine-grained channel state information (CSI) Use CSI to capture changes of multipath environment
Wall
Accesspoint
WallSmart
Appliance
Direct pathWiFi
deviceReflected
rays
Reflected rays
5 10 15 20 25 300
10
20
30
40
Subcarrier Index
SN
R
5 10 15 20 25 300
10
20
30
40
Subcarrier Index
SN
R
5 10 15 20 25 300
10
20
30
40
Subcarrier Index
SN
R
0 5000 100001280
1285
1290
1295
CSI.rate
Pac
ket
inde
x
CSI.rate
CS
I Am
plit
ude
...
Sub-carrier 1 Sub-carrier 2 Sub-carrier P-1 Sub-carrier P
Wall
Accesspoint
Wall
Direct pathWiFi
deviceReflected
rays
Reflected rays
5
Uniqueness of CSI Comparing to RSS
0
5
10
15
20
25Washing dishes
Talking on the phone
RS
S a
mpl
itude
0
5
10
15
20
25
30Washing dishes
Talking on the phone
CS
I Am
plitu
de
E-eyes System Challenges
Profile uniqueness and Robustness Generality to different types of activities Assisting the profile generation and updating
6
0 50 100 150 200 2500
5
10
15
20
25
30
Packet index
Sub
-car
riers
0 50 100 150 200 2500
5
10
15
20
25
30
Packet index
Sub
-car
riers
Activity Identification
Coarse Activity Determination Generality to different Activities
System Overview
7
Access Point Signal Time Series
Data Pre-processingIncrease robustness to real environments
Assisting the profile generation and updating
Profile Construction and Updating
User Feedback
Profile matching
None Profile Based
Construction
Adaptive
Updating
Walking Activity
Tracking using MD-
DTW
In-place Activity
Identification using EMD
Data FusionCrossing
Links
Known Activity Unknown Activity
Walking activity In-place activity
Coarse Activity Determination
8
Walking activity Large moving variance due to significant body movements and
location changes In-place activity
Small moving variance due to smaller body movements
… …
CS
I Am
pli
tud
e
Time
Subcarrier p
CS
I Am
pli
tud
e
Time
Subcarrier P
CS
I Am
pli
tud
e
Time
Subcarrier 1
)1(V … …)( pV )(PV
P
ppV
1)(V
CS
I Am
plit
ud
e
Time
In-place activity
Walking activity
Characteristics of CSI Measurements from Walking Activity
9
Trajectory 1 Trajectory 2
CSI pattern is dominated by walking activities’ path Doorway profile can facilitate walking activity tracking
0 50 100 150 200 2500
5
10
15
20
25
30
Packet index
Sub
-car
riers
0 50 100 150 200 2500
5
10
15
20
25
30
Packet index
Sub
-car
riers
Walking Activity Tracking
10
Walking ActivityClassifier
Multi-Dimensional Dynamic Time Warping Distance
Derivation
CSI measurements
DTW distance
Activity Profiles
CS
I Am
pli
tud
e
Time
Subcarrier P
CS
I Am
pli
tud
e
Time
Subcarrier 1
… …C
SI A
mp
litu
de
Time
Subcarrier p
CS
I Am
plit
ud
e
Time
Subcarrier P
CS
I Am
plit
ud
e
Time
Subcarrier 1
… …
CS
I Am
plit
ud
e
Time
Subcarrier p
Characteristics of CSI Measurements from In-Place Activity
11
Different in-place activities cause different distributions of CSI Different rounds of same in-place activities result in similar
distributions of CSI
0 2 4 6 8 10 12 14 16 18 200
5
10
15
CSI Amplitude Bins
Co
un
ts
0 2 4 6 8 10 12 14 16 18 200
5
10
15
CSI Amplitude Bins
Co
un
ts
Raw dataQuantized dataC
SI A
mp
litu
de
Time
In-place ActivityClassifier
CSI Amplitude BinsC
ou
nts
Distribution
In-Place Activity Identification
12
Distribution of CSI Amplitudes Extraction
Subcarrier Earth Mover’s Distance Derivation
CSI measurements
EMD distance
Activity Profiles
CSI measurements
CSI Amplitude Bins
Co
un
ts
Activity profile
CSI Amplitude Bins
Co
un
ts
Activity Identification
Profile Construction and Updating
Non-profiling Clustering
Semi-supervised approach to cluster daily activities and update CSI profiles
Construct CSI profiles when our system starts
13
Constructing profiles
Unknown Activity
Non-profiling Clusterin
gAdaptive Updating
User Feedback
Questions
How robust is the system in typical indoor environments?
Can two different activities be distinguished at the same location?
Is WiFi traffic in home environment feasible to identify activities?
14
Experimental Setup
WiFi devices Intel 5300 NIC + Thinkpad T500 and T 51 Cisco E2500
Scenarios Small apartment with one bedroom Large apartment with two beddoms
15
Questions
How robust is the system in typical indoor environments?
Can different activities be distinguished at the same location?
Is WiFi traffic in home environment feasible to identify activities?
16
Performance of In-place Activity Identification in Two Different Apartments
17
2-bedroom apartment
Activity typesT
PR
1-bedroom apartment
Activity types
TP
R
a b c d e f g o0
0.2
0.4
0.6
0.8
1
Single WiFi device Mult. WiFi device
a b f g h i j o0
0.2
0.4
0.6
0.8
1
Single WiFi device Mult. WiFi device
False positive rate: less than 5%
Performance of Walking Activity Tracking and Doorway Identification
18
1-bedroom apt. A B C D Unknown Door Door1 Door2 Door3
A 1 0 0 0 0Door1 1 0 0
B 0 1 0 0 0
C 0 0 0.95 0.05 0Door2 0 0.975 0.025
D 0 0 0 1 0
O 0 0 0.1 0 0.9 None 0 0 1
2-bedroom apt. E F G H Unknown Door Door1 Door2 Door3
E 1 0 0 0 0Door3 1 0 0
F 0.15 0.85 0 0 0
G 0 0 0.9 0.1 0Door4 0 0.875 0.125
H 0 0 0 1 0
O 0.05 0 0.1 0 0.9t None 0 0 1
Questions
How robust is the system in typical indoor environments?
Can different activities be distinguished at the same location?
Is WiFi traffic in home environment feasible to identify activities?
19
Performance of Identifying Different Activities at the Same Location
20
Four in-place activities Sleeping on the bed Sitting on the bed Receiving calls nearby the sink Washing dishes nearby the sink
4 5 6 7 8 9 10 11 120
0.20.40.60.8
1
True Positive Rate
Number of EMD bins
TP
R
4 5 6 7 8 9 10 11 120
0.10.20.30.40.5
False Positive Rate
Number of EMD bins
FP
R
Questions
How robust is the system in typical indoor environments?
Can different activities be distinguished at the same location?
Is WiFi traffic in home environment feasible to identify activities?
21
Performance of Different Packet Rate
22
Packet transmission rate (PTR): 5 pkts/s - 20 pkts/s
5 10 15 200.75
0.8
0.85
0.9
0.95
1
Average true positive rate
5 10 15 200
0.02
0.04
0.06
0.08
0.1
Average false positive rate
Conclusion
Show that the channel state information (CSI) from off-the-shelf 802.11n devices can be utilized to identify and distinguish in-place activities inside a home
Develop a monitoring framework that can run on a single WiFi AP and use the associated profile matching algorithms to compare amplitude profiles against those from known activities
Explore dynamic profile construction to accommodate the movement or replacement of wireless devices and day-to-day profile calibration
Extensive experiments in two apartments of different size demonstrates the generality of our approach
23
24