axe soota ubiquitous and mobile computing vishal rathi...

20
Ubiquitous and Mobile Computing CS 403x: Automatically Characterizing Places with Opportunistic CrowdSensing using Smartphones Ankit Kumar Vishal Rathi Axe Soota Computer Science Dept. Worcester Polytechnic Institute (WPI)

Upload: others

Post on 14-Oct-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Ubiquitous and Mobile Computing CS 403x: Automatically Characterizing

Places with Opportunistic CrowdSensing using Smartphones

Ankit Kumar Vishal Rathi

Axe Soota

Computer Science Dept.

Worcester Polytechnic Institute (WPI)

Page 2: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

● Traditional location sensing systems only make use of WiFi and GPS

● The error in GPS-, GSM-, or WiFi-based location estimates often ranges between 10 and 400 meters

● 426 of the 1,241 place visits incorrectly reported based on the location estimate

Problem Statement

Page 3: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Introducing CSP

● CSP - CrowdSense@Place● Interpretation of a location from Location Sensor

to user - as a place● Framework that exploits sensors that most

phone’s have● Smartly capture images and audio clips from

smartphones● Goal is to link place visits to various place

categories

Page 4: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Current Approach

● Place-discovery techniques these days:

● Exploit large-scale data collections, like point-of-interest databases (Google) to allocate place descriptors

Page 5: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Related Work

● Bing, Yelp

● Facebook, Twitter, FourSquare

● CenceMe - Similar application but doesn’t infer from images

● SenseCam - Goal to understand user’s environment

● VibN - Identifies points of interest in the city

Page 6: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

How is CSP different?

● CrowdSense@Place - Place classification based on existing methods to perform place segmentation

Page 7: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Overview

● Smartphone Application○ Sensing and Data Collection○ Privacy Settings

● Offline server-side processing○ Processing and Location Detection

Page 8: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Data Collection

● Audio detection○ “Do you have a Large size of these pants?”

● Pictures of objects

● Written Texts

Page 9: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Methodology

Page 10: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Smartphone Client

● Place Segmentation - WiFi fingerprinting and GPS to discover places

● Sensor Sampling - Simple heuristic to improve quality of data collected

● Privacy - Data resides on device for 24 hours

Page 11: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Sensor Data Classifiers

● Optical Character Recognition (OCR)● Indoor Scene Classification● Objects Recognition● Speech Recognition● Sound Classification

Page 12: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Place Modeling

● Data preprocessing○ Classifier Terms○ Mobility Terms

● Place Categorization

Page 13: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Results - Classifiers

● Indoor scene classification (GIST features) has the largest impact

● OCR does not have a strong overall effect● Object detection, speech recognition, and sound

classification had major effects

Page 14: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Results - Location Accuracy

● 69% Accuracy

● CSP outperforms GPS and Mobility by around 22% to 40%

● Mobility has 44% accuracy for workplace and 52% for college while CSP has 80% and 71% respectively

Page 15: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Applications of CSP

● Enhanced Local Search &

Recommendations

● Rich Crowdsourced Point-of-Interest

Category Maps

● Understanding City-scale Behavior Patterns

Page 16: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Limitations and Future Work

● Finer Place Categorization

● Privacy

● Activity vs. Place Category

● Energy Issues

Page 17: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Conclusions

● 36 person study

● Seven-weeks total

● 1241 places on 5 locations

● Average accuracy of 69%

Page 18: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

What we liked/disliked about the paper?

Likes:

● Graphs and tabulated data findings● The intensive study conducted● Limitations and issues considered

Dislikes:

● Doesn’t address privacy concerns appropriately

Page 19: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

Questions

Page 20: Axe Soota Ubiquitous and Mobile Computing Vishal Rathi …web.cs.wpi.edu/~emmanuel/courses/cs403x/D16/student... · 2016. 5. 13. · Ubiquitous and Mobile Computing CS 403x: Automatically

References

● http://www.fengzhao.com/pubs/ubicomp12_cps.pdf