rfid data management for pervasive computing applications · rfid is an increasingly ubiquitous...
TRANSCRIPT
![Page 1: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/1.jpg)
http://rfid.cs.washington.edu/The RFID Ecosystem Project
RFID Data Management for Pervasive Computing Applications
Evan Welbourne
University of Washington, CSE
EMC Innovation Conference 2008
Franklin, MA
![Page 2: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/2.jpg)
http://rfid.cs.washington.edu/
What is RFID?
Pervasive applications and challenges
The Cascadia system: RFID Data Manager
Handling holes in RFID data streams
Extracting high-level events from low-level RFID data
Supporting privacy
Facilitating application development
Outline
![Page 3: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/3.jpg)
http://rfid.cs.washington.edu/What is RFID?
Wireless ID and tracking
Captures information on:IdentityLocationTime
Unique identification
Passive (no batteries)
Reader
Tag
![Page 4: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/4.jpg)
http://rfid.cs.washington.edu/Passive RFID Markets
Today: $5 B industry, 2 billion tags
In 2018: $25 B industry, 600+ billion tags(source: IDTechEx)
![Page 5: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/5.jpg)
http://rfid.cs.washington.edu/
The RFID Ecosystem
100s of passive EPC Gen 2 tags
100s of RFID antennas
85,000 sq ft (8,000 sq m) building
Simulating an RFID-saturated future
![Page 6: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/6.jpg)
http://rfid.cs.washington.edu/
What is RFID?
Pervasive applications and challenges
The Cascadia system: RFID Data Manager
Handling holes in RFID data streams
Extracting high-level events from low-level RFID data
Supporting privacy
Facilitating application development
Outline
![Page 7: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/7.jpg)
http://rfid.cs.washington.edu/Pervasive RFID Applications
“Is my advisor in her office?”
“Where are my keys?”
Personal asset tracking
![Page 8: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/8.jpg)
http://rfid.cs.washington.edu/Pervasive RFID Applications
Personal asset tracking
Personal diary of physical events
“How did I spend my time?”
![Page 9: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/9.jpg)
http://rfid.cs.washington.edu/Pervasive RFID Applications
Personal asset tracking
Personal diary of physical events
Event-based search
“What emails did I send during the meeting a few weeks ago?”
![Page 10: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/10.jpg)
http://rfid.cs.washington.edu/Pervasive RFID Applications
Personal asset tracking
Personal diary of physical events
Event-based search
Focus on mobile and web 2.0 applications
![Page 11: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/11.jpg)
http://rfid.cs.washington.edu/Pervasive RFID Applications
Current trend: RFID in Hospitals
Tracking equipment, patients, personnel
Improved utilization of assets
We extend with more sophisticated inference
![Page 12: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/12.jpg)
http://rfid.cs.washington.edu/Application Challenges
Must cope with holes in RFID data stream
Must ensure a reasonable level or privacy
Application development is extremely difficult
The Cascadia system answers these challenges:
![Page 13: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/13.jpg)
http://rfid.cs.washington.edu/
What is RFID?
Pervasive applications and challenges
The Cascadia system: RFID Data Manager
Handling holes in RFID data streams
Extracting high-level events from low-level RFID data
Supporting privacy
Facilitating application development
Outline
![Page 14: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/14.jpg)
http://rfid.cs.washington.edu/Holes and Uncertainty in Data
Holes or gaps appear for two reasons:
Missed tag reads
![Page 15: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/15.jpg)
http://rfid.cs.washington.edu/Holes and Uncertainty in Data
Holes or gaps appear for two reasons:
Missed tag reads
Limited reader topology
![Page 16: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/16.jpg)
http://rfid.cs.washington.edu/Coping With Uncertainty
Smooth the data using a particle filterDirectly represent uncertainty with probabilistic data
Smoothed, Probabilistic Data
Smooth(ID, loc, time, prob)
Raw RFID Data
Raw(ID, loc, time)
ParticleFilter
Sensor
Model
Motion
Model
Connectivity
Graph(tag_1,(47.65,-122.3),2),
(tag_1,(47.65,-122.3),3), (tag_1,(47.65,-122.3),4), (tag_1,(47.65,-122.3),5),
...
(tag_1,’hall by 406’,3,.2), (tag_1, ’office 406’,3,.2), (tag_1, ‘office 408’,3,.2), (tag_1, ‘lab’,3,.2), (tag_1,‘hall by 408’,3,.2),
(tag_1, ’lab’,2,.33), (tag_1,’office 406’,2,.33), (tag_1,’office 408’,2,.33),
...
![Page 17: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/17.jpg)
http://rfid.cs.washington.edu/Coping With Uncertainty
Smooth the data using a particle filterDirectly represent uncertainty with probabilistic data
Smoothed, Probabilistic Data
Smooth(ID, loc, time, prob)
Raw RFID Data
Raw(ID, loc, time)
ParticleFilter
Sensor
Model
Motion
Model
Connectivity
Graph(tag_1,(47.65,-122.3),2),
(tag_1,(47.65,-122.3),3), (tag_1,(47.65,-122.3),4), (tag_1,(47.65,-122.3),5),
...
(tag_1,’hall by 406’,3,.2), (tag_1, ’office 406’,3,.2), (tag_1, ‘office 408’,3,.2), (tag_1, ‘lab’,3,.2), (tag_1,‘hall by 408’,3,.2),
(tag_1, ’lab’,2,.33), (tag_1,’office 406’,2,.33), (tag_1,’office 408’,2,.33),
...
Note on Data Rates / Volume
Average rates for 9am-5pm for 50 people, 300 tags:
Raw RFID data rate: 50 tag reads / secWith on-the-fly compression: 2 tag reads / sec
Raw data per week: 7,200,000 tag readsRaw data per week, compressed: 300,000 tag reads
Particle filter data rate: 500 particles / tag / sec
Smoothed data per week: 21B particles (20GB)
![Page 18: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/18.jpg)
http://rfid.cs.washington.edu/
What is RFID?
Pervasive applications and challenges
The Cascadia system: RFID Data Manager
Handling holes in RFID data streams
Extracting high-level events from low-level RFID data
Supporting privacy
Facilitating application development
Outline
![Page 19: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/19.jpg)
http://rfid.cs.washington.edu/Example: Higher-Level Event
Example Application: Track / verify patient treatmentsNurse is taggedEquipment is taggedDetect and log “treatment events”
ID location timeLocation Data
C D EBA
![Page 20: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/20.jpg)
http://rfid.cs.washington.edu/Example: Higher-Level Event
ID location timeNurse A 4
IV A 4
Location Data
C D EBA
Example Application: Track / verify patient treatmentsNurse is taggedEquipment is taggedDetect and log “treatment events”
![Page 21: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/21.jpg)
http://rfid.cs.washington.edu/Example: Higher-Level Event
ID location timeNurse A 4
IV A 4
Location Data
C D EBANurse B 5
IV B 5
Example Application: Track / verify patient treatmentsNurse is taggedEquipment is taggedDetect and log “treatment events”
![Page 22: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/22.jpg)
http://rfid.cs.washington.edu/Example: Higher-Level Event
ID location time
…
Nurse A 4IV A 4
Location Data
C D EBANurse B 5
IV B 5
Example Application: Track / verify patient treatmentsNurse is taggedEquipment is taggedDetect and log “treatment events”
![Page 23: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/23.jpg)
http://rfid.cs.washington.edu/Example: Higher-Level Event
C D EBA
Nurse treats patient in 407
Nurse B 5IV B 5
ID location time
…
Nurse A 4IV A 4
Location Data
Example Application: Track / verify patient treatmentsNurse is taggedEquipment is taggedDetect and log “treatment events”
![Page 24: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/24.jpg)
http://rfid.cs.washington.edu/
Translate to Event Specification Language
• PeexL –
“Probabilistic Event Extraction Language”
• SQL-like language
• SEQ sequencing construct
Extracting Events: Step 1
![Page 25: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/25.jpg)
http://rfid.cs.washington.edu/Extracting Events: Step 2
Extract Specified Events
PEEX – “Probabilistic Event Extractor”
Accepts specifications in PeexL
Periodically queries for new events
Assigns probabilities to extracted events
Events are stored or streamed to apps
Cascadia
Declarative Event Specifications
ProbabilisticEvent Streams
Raw RFID Data
Storage
Particle Filter
Smoothed
PEEXExtracted Events
![Page 26: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/26.jpg)
http://rfid.cs.washington.edu/Performance: Precision & Recall
Extracting a common event: ENTERED-MEETING-ROOM10 people, 30 tags, 30 minutes
Precision: Fraction of extracted events that match ground truth.Recall: Fraction of ground truth events captured by extracted events.
![Page 27: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/27.jpg)
http://rfid.cs.washington.edu/Performance: Precision & Recall
Extracting a common event: ENTERED-MEETING-ROOM10 people, 30 tags, 30 minutes
Precision: Fraction of extracted events that match ground truth.Recall: Fraction of ground truth events captured by extracted events.
All detected events Events which occur with certainty
25th
percentile
50th
percentile
75th
percentile
![Page 28: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/28.jpg)
http://rfid.cs.washington.edu/
1)
Precision/Recall trade-off: developer chooses best threshold
2)
Performance highly varied between tagsFor best tags at probability ≥ 0.3, Precision = .92, Recall = .85
3)
These results are for a very uncertain dataset:There are no readers inside rooms, only in hallways
![Page 29: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/29.jpg)
http://rfid.cs.washington.edu/
1)
Precision/Recall trade-off: developer chooses best threshold
2)
Performance highly varied between tagsFor best tags at probability ≥ 0.3, Precision = .92, Recall = .85
3)
These results are for a very uncertain dataset:There are no readers inside rooms, only in hallways
Bottom Line
Application only specifies event and probability threshold--
Cascadia does the rest
Ongoing improvement of event detection techniques!
![Page 30: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/30.jpg)
http://rfid.cs.washington.edu/
What is RFID?
Pervasive applications and challenges
The Cascadia system: RFID Data Manager
Handling holes in RFID data streams
Extracting high-level events from low-level RFID data
Supporting privacy
Facilitating application development
Outline
![Page 31: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/31.jpg)
http://rfid.cs.washington.edu/Default Policy: Physical Access Control
Problem: Asymmetric visibility among peers
“How do I know who can see my data?”
Concept:
Each user has a personal store of data / events
Stores only events that occurred when and where user was physically present
![Page 32: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/32.jpg)
http://rfid.cs.washington.edu/
sightings timestamp sightings timestamp sightings timestamp
Time: 0
’s data store ’s data store ’s data store
0 0 0
![Page 33: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/33.jpg)
http://rfid.cs.washington.edu/
sightings timestamp sightings timestamp sightings timestamp
Time:
’s data store ’s data store ’s data store
1 1 1
1
0 0 0
![Page 34: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/34.jpg)
http://rfid.cs.washington.edu/
sightings timestamp sightings timestamp sightings timestamp
Time:
’s data store ’s data store ’s data store
1 1 1
0 0 0
2 2 2
2
![Page 35: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/35.jpg)
http://rfid.cs.washington.edu/Default Policy: Physical Access Control
Key Points:Each user must carry a personal RFID badge
Enables applications that augment a user’s memory
Implemented as materialized authorization views in DB
Other “context-aware” policies are possible:“Only reveal my location during business hours”
“Only reveal my activity when I am in a meeting”
Interesting adaptations for probabilistic data
![Page 36: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/36.jpg)
http://rfid.cs.washington.edu/
What is RFID?
Pervasive applications and challenges
The Cascadia system: RFID Data Manager
Handling holes in RFID data streams
Extracting high-level events from low-level RFID data
Supporting privacy
Facilitating application development
Outline
![Page 37: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/37.jpg)
http://rfid.cs.washington.edu/Supporting Applications
Web 2.0 Personal Data Management Tool Suite:
Create/Manage Objects, Associate Tags
Specify Significant Places,
Share with Others
Specify Events,
Share with Others
Applications can integrate with tool suite
Declarative event language removes burden from developer
Event-based programming API simplifies development
![Page 38: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/38.jpg)
http://rfid.cs.washington.edu/Conclusions
RFID is an increasingly ubiquitous technology
More sophisticated applications face serious challenges:Uncertain data streams (gaps or holes in the data)Low-level data, high-level information needsPrivacy risksComplex application development
The Cascadia system addresses these challenges with:A probabilistic model of uncertaintyDeclarative event specificationsA probabilistic event extractorContext-aware access control policiesSuite of web-based personal data management tools
![Page 39: RFID Data Management for Pervasive Computing Applications · RFID is an increasingly ubiquitous technology More sophisticated applications face serious challenges: Uncertain data](https://reader034.vdocuments.net/reader034/viewer/2022052010/601fe2e8061a3420413ee856/html5/thumbnails/39.jpg)
http://rfid.cs.washington.edu/Thank You & Questions
Thank you!Misc notes:
See publications for details: http://rfid.cs.washington.edu/publications.html
Online demos available soon, see:http://rfid.cs.washington.edu/
Lots of ongoing and future work with Cascadia!