Lead Investigators

Jeff Burke, Mark Hansen, Deborah Estrin, Vern Paxson, Mark Allman

Overview

This project builds up best practices and reusable components by prototyping and executing specific data collection campaigns. We focus on the software that runs on the phones to collect the data, and the software used to manage and navigate the data during and after campaign execution. 

Approach

We usually have three campaigns in the pipeline either in planning & development, execution, or post-execution phase. For each campaign we have a team of people consisting of a campaign manager supported by a technical lead and a privacy auditor. The execution of each campaign results in data, a test of our software, and experience in managing the logistical and privacy requirements. We also learn about practical issues such as the extent to which an application may intrude into people's lives and daily routines, the circumstances under which they actually contribute data, etc. New campaigns typically involve a small group of participants, but will grow to include more participants as it matures, is debugged, and reworked.

Systems/Experiments

A collection data capture and upload software on the phone is developed or modified as required for each campaign. When faced with a large amount of data and management of several participants, we've found that there's much to be gained with a specific UI application for data management and navigation, which we are also beginning to accumulate. Our experiments include several data collection campaigns involving a combination of a common set of modalities -- location, image, audio -- but targeted for specific contexts, such as images during meal time, movement and travel in the first four hours of the day, coordinated recording of ambient noise in a specific area of campus, etc.

Accomplishments

As mentioned earlier, we've executed several data collection campaigns resulting in useful data. We collected three days (8 hours each) of images collected about once a minute from Sensys 2006 with five participants; We have gathered two weeks of GPS data for four hours a day from a dozen participants; We've collected images of dozens of meals over two weeks for a diet monitoring campaign.

The campaigns have also helped us exercise and debugge data capture and upload software for the Symbian S60 phones. With tools written in Python and Java, we have a growing collection of software to gather GPS data, cell tower ID, image, and audio data. The measurements are opportunistically uploaded to sensorbase, providing "live" feedback of the campaign's status to the campaign manager.

For data browsing and navigation, we've developed a web-based data browsing front end called Imagescape for GPS, image, and audio data. The user is able to sort of filter the data using a map for location attributes, power measurements for audio data, and dominant color for image data. We've also developed an image clustering algorithm to help the user browse the images after being partitioned into sequences of images with similar color histograms. The implementation of Imagescape required backed server development that acts as an intermediary between Imagescape and sensorbase in order to carry out the necessary image and audio processing used for the filtering and clustering of images.

Future Directions

A top concern of users is the security and confidentiality of the data collected about them. As a future direction, we are planning work for the development of a sharing policy language, authoring and visualization tools, and selective sharing services (data scrubbers). For Imagescape, we plan to improve interoperation with sensorbase. We are also interested in tools for managing campaigns and visualizing campaign data. Finally, for the phone, we are working on an application that can read execute campaign descriptions, rather than creating custom applications for each campaign.

People

Faculty:

• Jeff Burke, UCLA, School of Film, Theater, and Television
• Mark Hansen, UCLA, Department of Statistics
• Deborah Estrin, UCLA, Computer Science Department

Researchers:

• Vern Paxson, ICSI Center for Internet Research
• Mark Allman, ICSI Center for Internet Research

Staff:

• Nicolai Petersen, UCLA, Computer Science Department

Graduate Students:

• Andrew Parker, UCLA, Computer Science Department
• Sasank Reddy, UCLA, Electrical Engineering Department
• August Joki, UCLA, Computer Science Department
• Eitan Mendelowitz, UCLA, Computer Science Department
• Gong Chen, UCLA, Department of Statistics
• Thomas Schmid, UCLA, Electrical Engineering Department
• Vidyut Samanta, UCLA, Computer Science Department

Undergraduates:

• Frank Chen, UCLA, Computer Science Department
• Elisa Yuen, UCLA, Electrical Engineering Department
• Jason Ryder, UCLA, Computer Science Department
• Stephen Oakley, UCLA, Computer Science Department
• Robert Wei, UCLA, Computer Science Department
• Eric Chang, UCLA, Computer Science Department

External Research Partnerships

Nokia (present)
ICSI (present)
Cisco (present)