Location Without GPS John Krumm Microsoft Research Redmond, Washington, USA.

Location Without GPS

John KrummMicrosoft Research

Redmond, Washington, USA

Seattle, Washington, USA

Kyoto, Japan

Location

Importance of Location• Find your way• Find nearby things• Invoke location-based

services– Electronic graffiti, e.g. “There is

a better Mexican restaurant 0.2 km north of here.”

– List of nearby events

• Part of context– In lecture hall → cell phone off– At home → use home network

IWMMS & Location“Study of Structuring and Recalling Life Log Experience Using Location Information”, Y. Aihara, R. Ueoka, K. Hirota and M. Hirose

-- Already using location for activity inference

“Active Wearable Vision Sensor: Recognition of Human Activities”, K. Sumi, M. Toda, S. Tsukizawa and T. Matsuyama

“Cooperative Dialogue Planning with User and Situation Models via Example-based Training”, I. R. Lane, S. Ueno and T. Kawahara

-- Inferring context of user – location is part of context

“A Hybrid Dynamical System for Event Segmentation, Learning, and Recognition”, H. Kawashima, K. Tsutsumi and T. Matsuyama

“Time-Series Human-Motion Analysis with Kernels derived from Learned Switching Linear Dynamics”, T. Mori, M. Shimosaka, T. Harada and T. Sato

-- Apply HDS/SLDS to infer location & mode of transportation & destination?

Why Not Use GPS?

• Does not work indoors• Needs view of satellites

Location Sensing

Hazas, Scott, Krumm, “Location-Aware Computing”, IEEE Computer Magazine, February 2004.

Outline

• Introduction

• LOCADIO – Wi-Fi triangulation

• NearMe – Wi-Fi proximity

• RightSPOT – FM radio triangulation

• TempIO – Inside/outside from temperature

Location from 802.11 with LOCADIO*

• Mobile device measures signal strengths from Wi-Fi access points• Computes its own location

Wi-Fi (802.11) access point

* Location from Radio

with Eric Horvitz

LOCADIO – Radio Survey

Radio survey to get signal strength as a function of position

LOCADIO - Constraints

No passing through walls No speeding

200 400 600 800 1000 1200

still

moving

still

moving

still

moving

A Posteriori Probability of Move

time (seconds)

actual

unsmoothed

smoothed

We know when you move

Make the client as smart as possible to reduce calibration effort

LOCADIO - Results

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 160

0.1

0.2

0.3

error in meters

rela

tive

fre

qu

ency

Error Histogram

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 160

0.5

1

error in meters

rela

tive

fre

qu

ency

Cumulative Error Distribution

Hidden Markov model gives median error of 1.53 meters

Outline

• Introduction

• LOCADIO – Wi-Fi triangulation

• NearMe – Wi-Fi proximity

• RightSPOT – FM radio triangulation

• TempIO – Inside/outside from temperature

NearMe

Person

conferencerooms

printers

bathroom reception desk

people

Find people and things nearby

Download from http://research.microsoft.com/~jckrumm/NearMe.htm

with Ken Hinckley

The Basic Idea802.11 Wi-Fi access point

NearMe Proximity Server

Download from http://research.microsoft.com/~jckrumm/NearMe.htm

Location vs. Proximity

s1 = measured signals

s2 = measured signals

x1 = (x,y) location

x2 = (x,y) location

d12 = f(x1, x2)

d12 = g(s1, s2)

Download from http://research.microsoft.com/~jckrumm/NearMe.htm

NearMe Client

Windows XP PocketPC 2003

Requirements:• Windows XP• WWW access• Microsoft .NET Framework

Download from http://research.microsoft.com/~jckrumm/NearMe.htm

NearMe Client – Test Connections

Download from http://research.microsoft.com/~jckrumm/NearMe.htm

NearMe Client – Register

Register with:• Name• Email (optional)• URL (optional)• Expiration interval

Download from http://research.microsoft.com/~jckrumm/NearMe.htm

NearMe Client – Report Wi-Fi

• List of detectable Wi-Fi access points• Access points used only as beacons

• Periodic reports for mobility

Download from http://research.microsoft.com/~jckrumm/NearMe.htm

NearMe Client -- Query

Adjustable “Look back” time to filter outdated reports

Download from http://research.microsoft.com/~jckrumm/NearMe.htm

Register as thing

NearMe Client – Nearby Things

person elevator kitchen bathroom

conference room stairs mail room stitchable device

printer cafeteria reception desk demo person

Report signal strengths Query for things

Download from http://research.microsoft.com/~jckrumm/NearMe.htm

Simple Distance Function

1

2

3

4

5 -1 -0.75 -0.5 -0.25 0.0 0.25 0.5 0.75 1.0

5

10

15

20

25

Number of Access Points in Common, n∩

Spearman Correlation, ρs

d = -2.53∙n∩ – 2.90∙ρs - 22.31

rms error = 14.04 meters

ρs = 0.39

Download from http://research.microsoft.com/~jckrumm/NearMe.htm

Access Point Layout

21

3

AF

B

D

C

E

F

A B C D E

• Access point topology in database• Recomputed every hour

Download from http://research.microsoft.com/~jckrumm/NearMe.htm

NearMe Demo

Outline

• Introduction

• LOCADIO – Wi-Fi triangulation

• NearMe – Wi-Fi proximity

• RightSPOT – FM radio triangulation

• TempIO – Inside/outside from temperature

SPOT Watch Location

weather traffic

dining movies

Commercial FM: transmit new data every ~2 minutes

Filter on watch to take what it wants

Watch displays “personalized” data

with Adel Youssef, Ed Miller, Gerry Cermak, Eric Horvitz

Location-Sensitive Features

Nice to have

• Local traffic• Nearby movie times• Nearby restaurants

Need to know location of device …

Use FM Radio Signal StrengthsKPLU 88.5KAOS 89.3KASB 89.3KNHC 89.5KWFJ 89.7KGHP 89.9KGRG 89.9KUPS 90.1KEXP 90.3KSER 90.7KVTI 90.9KBCS 91.3KBTC 91.7KLSY 92.5KUBE 93.3KMPS 94.1KRXY 94.5KUOW 94.9KJR 95.7KYPT 96.5KBSG 97.3KING 98.1KWJZ 98.9KISW 99.9KQBZ 100.7KPLZ 101.5KZOK 102.5KMTT 103.7KMIH 104.5KFNK 104.9KCMS 105.3KBKS 106.1KRWM 106.9KNDD 107.7

Scan signal strengths of 32 FM radio stations at 1 Hz

0 100 200 300 400 500 600 700560

570

580

590

600

610

620

630

640

650

660

Sample Number

RS

SI

RSSI Values in One Suburb

raw rssimedian filtered

Ranking Approach

• Each watch scales signal strengths differently• Impractical to calibrate every watch

Input Power

Mea

sure

d P

ower

A B C

Redmond: KPLU < KMTT < KMPSBellevue: KMTT < KPLU < KMPSIssaquah: KMTT < KMPS < KPLU…

Any monotonically increasing function of signal strength preserves ranking

N radio stations → N! possible rankings

1. A B C2. A C B3. B A C4. B C A5. C A B6. C B A

Test

RedmondWoodinville

KirklandBellevue

IssaquahSammamish

01

23

45

0

0.2

0.4

0.6

0.8

1

LocationPermutation Hash Code

Six suburbs and six radio stations

81.7% correct from 8 radio stations

Avoid Manual Training

Seattle KMPS 94.1 MHz KSER 90.7 MHz

Classify Into Grid Cell

• Find location in grid

• Use predicted signal strengths to avoid manual training

≈ 8 kilometers average error

Summer intern Adel Youssef, U. Maryland

Outline

• Introduction

• LOCADIO – Wi-Fi triangulation

• NearMe – Wi-Fi proximity

• RightSPOT – FM radio triangulation

• TempIO – Inside/outside from temperature

TempIO – Inside/Outside ClassificationSuunto X9 – GPS, altimeter, thermometer

Suunto N3 – SPOT watch, knows outside temperature, location

Are you inside or outside?

• Turn off GPS if inside to save batteries• Metadata for digital photos • Higher-level context reasoning

Inside/Outside Outside Temperature Inside Temperature

Measured Temperature

Outside

Inside

Measured

Bayes Net

with Ramaswamy Hariharan

World Weather Stations

6509 weather stations → http://weather.noaa.gov/weather/metar.shtml → our web service

Inside/Outside from Temperature

Inside/Outside Accuracy

00.10.20.30.40.50.60.70.80.9

1

Seattle, WA Barrow, AK Key West,FL

San Diego,CA

Atlantic City,NJ

Fra

ctio

n C

orr

ect

in 2

003

• From hourly temperature data in five US cities, 2003• Average correct 81%

Kyoto

The End