Automatically Characterizing Places with Opportunistic CrowdSensing using Smartphones
1 like277 views
This document summarizes research on automatically characterizing places using opportunistic crowdsensing with smartphones. A framework called CrowdSense@Place uses sensors like GPS, WiFi, microphones and cameras on smartphones to collect location data and audio/visual hints about places. These hints are classified to identify place categories like restaurants, shops etc. Evaluations with 36 users across 1300 places achieved 69% accuracy in place categorization. While effective for analyzing large-scale patterns, limitations include slow learning rates, energy usage and privacy concerns when collecting and sharing sensor data.
Automatically Characterizing Places with Opportunistic CrowdSensing using Smartphones
- 2. Automatically Characterizing Places with Opportunistic CrowdSensing using Smartphones UbiComp’12, Pittsburgh, USA Best Paper Award Authors: Yohan Chon Nicholas D. Lane Fan Li Hojung Cha Feng Zhao
- 3. Characterizing Places Legend: Educational Institutions Restaurants Hospitals Shopping Marine Drive, Mumbai, India
- 5. CrowdSense@Place (CSP) Categorizes places Logical location meaningful to user Links places with Place categories Grocery store, restaurant, hospital, university Activity Shopping, eating, working
- 7. Collecting Data: How? Location and user trajectories using Wi‐Fi/GPS Samples data from sensors Microphone Camera Crowdsourcing Collect large volumes of data
- 8. Collecting Data: What? Audio and visual place hints mined from opportunistic sensor data Spoken words “Can I have a Cappuccino please?” Physical objects Cups, shoes, clothes Written texts Menu, posters, hoardings
- 9. Collecting Data: When? User uses phone Calls, emails, or browses Concern: Privacy Full control of data collection Buffered before transmission Review collected data Option to delete before upload
- 11. Extracting Hints Image and audio classifiers Scene classification Object recognition Optical character recognition Speech recognition Sound recognition Output merged with location based signals Wi‐Fi, GPS
- 13. Let’s Try To Pick Up Hints The Coffee Bean Order Here Can I get a Latte to go please? The Coffee Bean, India
- 14. Let’s Try To Pick Up Hints Laptop Dell SSD Store
- 15. CSP Working Place as a document Builds the document with sensor based hints ID: WiFi Fingerprint Bloomingdale (0.75) mannequin (0.87) trouser (0.83) blouse (0.65) shirt (0.76) belt (0.4) bag (0.56) outlet (0.87) store (0.76) 35‐75% (0.23)
- 16. CSP Framework
- 17. Opportunistic Sensing of Data Smartphone Application usage Phone calls, browsing Piggy‐back on user actions Screen state and light sensor Accelerometer Orientation, movement GPS & Wi‐Fi Microphone Camera
- 18. Sensor Data Classifier Object Detection Optical Character Recognition Sound Classification Speech Recognition
- 19. Sensor Data Classifier Hints v/s Noise Filter out the data Phone is shaky or facing down Crowdsourcing Repeated visits to place
- 20. CSP Framework
- 21. Applications Location based reminders Content Delivery Activity recognition Understanding City‐Scale Patterns Enhanced Local Search & Recommendations Awareness of the types of places a user frequently visits leading to additional user profile attribute Rich CrowdSourced Point‐of‐Interest Category Maps Maps that relate places to place categories A targeted advertising app
- 23. Evaluations Statics: 36 users 5 locations 1241 places 1,300 places 46,000 hours 2,300 images 4,200 audios 22% of images are either blurred or completely black Accuracy: 69%
- 24. Evaluations Questions How accurate? Which features types are most discriminative? How well do certain feature types operate in noisy environments?
- 25. Evaluations Categories College & Education, Arts & Entertainment, Food & Restaurant, Home, Shops, Workplace, Others Metrics Accuracy of place categorization: (No of correctly recognized places)/(No of places evaluated)
- 26. Evaluations
- 27. Conclusion Efficient categorization of places Uses hints, like humans do Effective use of crowd sensing Accurate classifier Advanced applications Large scale evaluations Power consumption Privacy concern Future, User participation Social Networking Sites
- 28. References http://www.msr‐waypoint.com/en‐ us/um/people/zhao/pubs/ubicomp12_cps.pdf D. Ashbrook and T. Starner. Using GPS to Learn Significant Locations and Predict Movement Across multiple users. http://foursquare.com
- 29. Questions