SCPL: Indoor Device-Free Multi-Subject Counting and Localization...

WINLAB

SCPL: Indoor Device-Free Multi-Subject Counting and Localization Using

Radio Signal Strength

Rutgers University

Chenren Xu

Joint work with Bernhard Firner, Robert S. Moore, Yanyong Zhang

Wade Trappe, Richard Howard, Feixiong Zhang, Ning An

WINLAB

Device-free Localization

2

WINLAB

Device-free Localization

3

WINLAB

Why Device-free Localization?

q  Monitor indoor human mobility

4

WINLAB

Why Device-free Localization?

q  Monitor indoor human mobility

Elder/health care 5

WINLAB

Why Device-free Localization?

q  Monitor indoor human mobility

Traffic flow statistics 6

WINLAB

Why Device-free Localization?

q  Monitor indoor human mobility

Traffic flow statistics 7

WINLAB

Why Device-free Localization?

q  Monitor indoor human mobility

q Health/elder care, safety

q Detect traffic flow

q  Provides privacy protection

q No identification

8

WINLAB

Why Device-free Localization?

q  Monitor indoor human mobility

q Health/elder care, safety

q Detect traffic flow

q  Provides privacy protection

q No identification

q  Use existing wireless infrastructure

9

WINLAB

Previous Work

q  Single subject localization

q Fingerprinting-based approach

10

WINLAB

Fingerprinting N Subjects

q  Multiple subjects localization

q Needs to take calibration data from N people

for localizing N people

11

WINLAB

Fingerprinting N Subjects

12

…

9 trials in total for 1 person

WINLAB

Fingerprinting N Subjects

13

…

WINLAB

Fingerprinting N Subjects

14

…

…

WINLAB

Fingerprinting N Subjects

15

…

…

36 trials in total for 2 people!

WINLAB

Fingerprinting N Subjects

16

1 person

9 cells 9

9 × 1 min = 9 min

WINLAB

Fingerprinting N Subjects

17

1 person 2 people

9 cells 9 36

36 cells 36 630

630 × 1 min = 10.5 hr

WINLAB

Fingerprinting N Subjects

18

1 person 2 people 3 people

9 cells 9 36 84

36 cells 36 630 7140

100 cells 100 4950 161700

161700 × 1 min = 112 days The calibration effort is prohibitive !

WINLAB

SCPL q  Input

q  Collecting calibration data only from 1 subject (D1)

q  Observed RSSI change caused by n subjects

q  Output

q  count and localize N subjects.

q  Main Insight:

q  If the number n is known, localizing n subjects

is strightforward 19

WINLAB

No Subjects

20

WINLAB

One Subject

21

WINLAB

Two Subjects

22

WINLAB

Measurement

23

N = 0 N = 1 N = 2 Link 1 0 4 4 Link 2 0 5 7 Link 3 0 0 5 Total (∆N) 0 9 16

∆N N?

∆N / ∆1 = N?

WINLAB

Measurement

24

WINLAB

Measurement

25

Nonlinear problem!

1.6

∆N / ∆1 ≠ N

WINLAB

Measurement

26

4 dB

5 dB

WINLAB

Measurement

27

5 dB

6 dB

WINLAB

Measurement

28

4 dB + 0 dB = 4 dB √ 5 dB + 6 dB = 11 dB ≠ 7 dB X 0 dB + 5 dB = 5 dB √

= + ?

4 dB

5 dB

5 dB

6 dB

5 dB

7 dB

4 dB

WINLAB

Measurement

29

5 dB + 6 dB ≠ 7 dB X

Shared links observe nonlinear fading effect from multiple people

≠ + ! 5 dB 6 dB 7 dB

WINLAB

SCPL Part I Sequential Counting (SC)

30

WINLAB

Counting algorithm

31

WINLAB

Phase 1: Detection

32

Measurement in 1st round

5 dB

7 dB

4 dB ∆N = 4 + 7 + 5 = 16 dB

∆N > ∆1

Subject Count ++

WINLAB

Phase 2: Localization

33

Measurement in 1st round

5 dB

7 dB

4 dB

Find this guy

PC-DfP:

C. Xu, B. Firner, Y. Zhang, R. Howard, J. Li, and X. Lin. Improving rf-based device-free passive localization in cluttered indoor environments through probabilistic classification methods. In Proceedings of the 11th international conference on Information Processing in Sensor Networks, IPSN ’12

WINLAB

Phase 3: Subtraction

34

Calibration data

5 dB

6 dB

WINLAB

Phase 3: Subtraction

35

Subject count ++ Go to the next iteration…

= -

Measurement in 1st round

Calibration data

Measurement In 2nd round

4 dB

1 dB

5 dB

6 dB

5 dB

7 dB

4 dB

WINLAB

Phase 3: Subtraction

36

Subject count ++ Go to the next iteration…

Hold on …

= -

Calibration data

Measurement in 1st round

Measurement In 2nd round

4 dB

1 dB

5 dB

6 dB

5 dB

7 dB

4 dB

WINLAB

Phase 3: Subtraction

37

Measurement In 2nd round

4 dB

1 dB

WINLAB

Phase 3: Subtraction

38

Measurement In 2nd round

Calibration data

4 dB

1 dB

4 dB

5 dB

WINLAB

Phase 3: Subtraction

39

Measurement In 2nd round

Calibration data

= -

We over-subtracted its impact on shared link!

4 dB

1 dB

4 dB

5 dB -4 dB

WINLAB

Measurement

40

WINLAB

Measurement

41

1st round

WINLAB

Measurement

42

1st round

WINLAB

Measurement

43

1st round

2st round

WINLAB

Phase 3: Subtraction

44

We need to multiply a coefficient β ϵ [0, 1] when subtracting each link

= -

Calibration data

Measurement in 1st round

Measurement In 2nd round

4 dB

1 dB

5 dB

6 dB

5 dB

7 dB

4 dB

WINLAB

Location-Link Correlation

q  To mitigate the error caused by this over-

subtraction problem, we propose to

multiply a location-link correlation

coefficient before successive subtracting:

45

WINLAB

Phase 3: Subtraction

46

Subject count ++ Go to the next iteration…

= - 4.6 dB 6 × 0.4 dB

Measurement in 1st round

Calibration Data

Measurement in 2nd round

5 × 0.8 dB

4 dB

5 dB

7 dB

4 dB

1 dB

WINLAB

Phase 3: Subtraction

47

Measurement in 2nd round

Calibration data

= -

Measurement in 3rd round

We are done !

4.6 dB

4 dB

1 dB

6 × 0.6 dB

4 × 0.8 dB 1 dB

1 dB

1 dB

WINLAB

SCPL Part II Parallel Localization (PL)

48

WINLAB

Localization

q  Cell-based localization

q Allows use of context information

q Reduce calibration overhead

q Classification problem formulation

49

C. Xu, B. Firner, Y. Zhang, R. Howard, J. Li, and X. Lin. Improving rf-based device-free passive localization in cluttered indoor environments through probabilistic classification methods. In Proceedings of the 11th international conference on Information Processing in Sensor Networks, IPSN ’12

WINLAB

Linear Discriminant Analysis q  RSS measurements with person’s presence in each

cell is treated as a class/state k

q  Each class k is Multivariate Gaussian with common

covariance

q  Linear discriminant function:

50

Link 1 RSS (dBm) L

ink

2 R

SS (d

Bm

) k = 1 k = 2

k = 3

WINLAB

Localization

q  Cell-based localization

q  Trajectory-assisted localization

q  Improve accuracy by using human mobility

constraints

51

WINLAB

Human Mobility Constraints

52

You are free to go anywhere with limited step size inside a ring in free space

WINLAB

Human Mobility Constraints

53

In a building, your next step is constrained by cubicles, walls, etc.

WINLAB

Phase 1: Data Likelihood Map

54

WINLAB

Impossible movements

55

WINLAB

Impossible movements

56

WINLAB

Phase 2: Trajectory Ring Filter

57

WINLAB

Phase 3: Refinement

58

WINLAB

Here you are!

59

WINLAB

Viterbi optimal trajectory

q  Single subject localization

q  Multiple subjects localization

60

ViterbiScore =

WINLAB

System Description

q  Hardware: PIP tag

q Microprocessor: C8051F321

q Radio chip: CC1100

q Power: Lithium coin cell battery

q  Protocol: Unidirectional heartbeat (Uni-HB)

q Packet size: 10 bytes

q Beacon interval: 100 msec 61

WINLAB

Office deployment

62 Total Size: 10 × 15 m

WINLAB

Office deployment

63 37 cells of cubicles, aisle segments

WINLAB

Office deployment

64 13 transmitters and 9 receivers

WINLAB

Office deployment

65 Four subjects’ testing paths

WINLAB

Counting results

66

WINLAB

Counting results

67

WINLAB

Localization results

68

WINLAB

Open floor deployment

69 Total Size: 20 × 20 m

WINLAB

Open floor deployment

70 56 cells, 12 transmitters and 8 receivers

WINLAB

Open floor deployment

71 Four subjects’ testing paths

WINLAB

Counting results

72

WINLAB

Localization results

73

WINLAB

Conclusion and Future Work q  Conclusion

q  Calibration data collected from one subject can be used to

count and localize multiple subjects.

q  Though indoor spaces have complex radio propagation

characteristics, the increased mobility constraints can be

leveraged to improve accuracy.

q  Future work

q  Count and localize more than 4 subjects 74

WINLAB

Q & A

Thank you

75