Transferring Positioning Model for Device-free Passive Indoor - - PowerPoint PPT Presentation

Transferring Positioning Model for Device-free Passive Indoor Localization

Kazuya Ohara* ，Takuya Maekawa （Graduate School of Information Science and Technology, Osaka University） Yasue Kishino, Yoshinari Shirai, Futoshi Naya （NTT Communication Science Laboratories） 1

Research Background

Device-free passive indoor positioning 2

• Estimate position of person

in indoor environment

• Need not to carry any device
• surveillance of elderly person
• smart homes automation
• R1

human body interferes Positioning Model

RSSI

R2

R1 R2 RSSI R1 R2

𝑦, 𝑧

RSSI R1 R2

𝑦, 𝑧

RSSI R1 R2

𝑦, 𝑧 Wi-Fi AP Receiver

Research Purpose

Issue

Collecting labeled training data at many positions in the target environment is costly

Purpose Construct an indoor positioning model by transferring training

data from other environments (source environments)

3

AP Receiver between AP and receiver Training points

Purpose Issue transfer

source environment target environment

Approach

4

0.5 1 1.5 2

• 54
• 53
• 52
• 51
• 50
• 49

2 4 6 8 10

Variance RSSI[dBm] Time[s]

raw signal strength variance of signal strength to track the person

raw RSSI variance Variance Position

variance model

transfer Variance Position

source environment target environment AP Receiver AP Receiver

Overview of proposed method

5

Learning variance model in source environments

Floor plan

(x1,y1) (x1,y1)

Wi-Fi data

Learn variance model Learning positioning model for target environment

Floor plan of target env.

Transfer variance model Learn models

Positioning model

6

t t t t

？

RSSI RSSI RSSI RSSI

Proposed method (Necessary information)

source environment target environment

• Floor plans (layout of walls)
• RSSI when there is no person
• Labeled RSSI

e.g. by video recording

• Floor plans (layout of walls)
• RSSI when there is no person
• Unlabeled RSSI

Learning variance model

0.5m 0.5m

Example data

Variance

signal strength variance values when the person passed each position

AP Receiver ⋯

7

Passing position

Bimodal model

• Largest at AP and Receiver
• mixture of 2 Gaussian functions

𝑤𝑦

𝑦

Variance model Become larger as the path becomes closer to device Human body interferes signal more 𝑦 : passing point, 𝑤𝑦 : variance value

Learning variance model

Example data signal strength variance values when the person passed each position

AP Receiver Wall

Variance

0.5m 0.5m⋯

8

Variance model

variance Position

Passing position Variance model

variance Position

Variance model Feature depends greatly on the area divided by walls sub-line segment

Transferring variance model

• 1. Find top-k similar source sub-line segments

9 i. the same end points of sub-line segment ii. the same number of walls

• iii. kNN search according to three criteria
• 2. Transferring variance model

target sub-line segment source environment

Averaging parameters of top-k source models target model parameter

Transferring variance model

10

2m 3m

RSSI Probability target source

materials on segments are similar

• 1. Length of sub-line segment
• 2. Signal strength when there is no person

compare distribution Criteria used for selecting source sub-line segments with kNN search

• 3. Variance value when a person passes randomly

Variance Frequency Variance model

variance Position

Variance Frequency

variance Position

similar

Transferring variance model

11

2m 3m

RSSI Probability target source

materials on segments are similar

• 1. Length of sub-line segment
• 2. Signal strength when there is no person

compare distribution Criteria used for selecting source sub-line segments with kNN search

Variance Time Variance

• 3. Variance value when a person passes randomly

Variance Time Probability target source

compare distribution From unlabeled data

• utlier detection

Learning models

12

Passing detection model

• Detecting whether or not a person passes
• Training data : variance values observed

by selected source sub-line segments

Variance SVM Hyperplane Variance passing non-passing

Learning models

13

Positioning model

• Estimate position of a person when passing is detected
• Compute variance values at some points from a transferred

variance model

Variance Variance Variance Variance Variance Variance model

variance Position

Two positions match Tracking by using particle filter

Tracking by using particle filter

14

t-1 t t

• Particle : position of a person
• High density particles

High probability that the person is there Particle filter t-2

Evaluation (Environments)

• 1 Wi-Fi AP and 10 receivers
• Walk around for 20 minutes in each environment
• Leave-one-out cross-validation evaluation

11.7m 16.3m env.1 8.4m 12.9m env.2 10.4m 12.3m env.3 7.9m 10.4m env.4 AP Receiver

15

Evaluation (Results)

16

Error distance [m]

Supervised・・・trained on labeled data obtained in the same environment Random・・・select randomly k sub-line segments while transferring variance model

• Performance of random was poorer

than with proposed method Evaluating proposed method

Criteria

Criteria for selecting source sub-line segments

• “Length” contribute more than other criteria
• “all” achieve best performance

L ： Length of sub-line segment S ： Signal strength when there is no person V ： Variance value when a person passes randomly

1.5 1.6 1.7 1.8 1.9 2.0 all LS LV SV L S V

Error distance [m]

1.63 1.71 2.35 1.0 1.5 2.0 2.5

Supervised Proposed Random 0.08m

Conclusion

• We proposed a new method that enables us to

construct a positioning model for device-free passive indoor localization with little effort

• As a part of our future work, we plan to automatically
• btain unlabeled data in an end user’s daily life to

reduce burdens imposed on the user

17