Download pdf - WLCD: A Web Lecture Capture System with Robotic …atlascollab.umich.edu/docs/CHEP2006_WLCD-Talk.pdf · Capture System with Robotic Speaker Tracking. Mumbai, ... • CERN Training

Steven Goldfarb for Jeremy HerrCHEP

Mumbai, India - 15 Feb 2006

WLCD: A Web LectureCapture System with Robotic

Speaker Tracking

Mumbai, India - 15 Feb 2006 J. Herr - University of Michigan WLCD - Slide 2

Partnership University of Michigan

• UM ATLAS Collaboratory Project• Department of Physics• Media Union

CERN• Technical & Acedemic Training• Summer Student Program• IT

American Physical SocietyPrimary Goals

implement an electronic archival system for slide-based presentations onthe Internet;

assess the ability of the system to improve upon and complement existingarchival methods;

determine the resources required to install and maintain the system as astandard archival service.

The Web Lecture Archive Project

Web Lecture Capture Device


Web Lecture Recording ATLAS & LHC In General

• Physics & Software Tutorials• ATLAS Week plenary sessions• CERN Training Seminars, Tutorials

American Physical Society• Opportunities for Physics in Biology 2002, 2004• April Meeting 2004

Fermilab• Software Tutorials

University of Michigan• Saturday Morning Physics• Medical School Symposium on Clinical Research 2004

Harvard• International Conference for Systems Biology 2005

Current Recording Activities


Motivations Driving Automation Simultaneous Recording of Multiple Lectures Limited Manpower Need for Instant Publication

Components of Lecture Archiving Requiring Automation Video Capture and Compression Audio Recording Slide Capture and Synchronization Construction, Publication of Electronic Lecture

Development of Automated Recording Technology


Brief History of Lecture Recording Automation 1999: Used Sync-O-Matic 3000 Software to Record CERN Summer School Lectures

• Sync-O-Matic completely automatic if Installed on presentation computer Slides in Powerpoint

• But Everybody wrote plastic transparencies back then! And access to presentation computers not always possible

• So Had to record timing by hand, process lectures after the talk.

2000: Extra video camera was pointed at the screen• timing information was collected manually afterward

2001: Giosue Vitaglione wrote CarpePPT• Same idea as Sync-O-Matic, but producing Lecture Objects (non-proorietary)

2002: Jake Bourjaily wrote CarpeLecture• Extension, allowing PDF and integrated with CDS Agenda

Now: Web Lecture Capture Device• VGA splitter connected to signal coming from presentation computer• high-resolution capture card is used with change-detection software

captures every image displayed on the speaker’s computer No limit to format (PowerPoint, PDF, Web pages, command line prompts,…)

Development of Automated Recording Technology (2)


Lecture Object Standardize Storage of Web Lecture Ingredients, Recipe

• Archive media in standard formats (mpeg-4, jpeg, etc.)• XML extension to SMIL (Synchronized Multimedia Integration Language)

Describes slide timing, media location, other information SMIL = W3C Standard

• Add metadata with RDF (Resource Description Framework) RDF = W3C Standard

Standard Procedure for WLAP

Recall the Lecture Object (Talk before Coffee)

mpeg-4

jpeg

XMLDescription

meta data

Lecture Object

Transformations


Lecture Object - Definition

Organized Directory Structure and Naming System

ContentsAudio/Video media: MPEG-4 video fileSlides: JPEG images of everything displayed on lecturer’s screenSupporting Documents: original PowerPoint file, animations, etc.Log Files: describing technical history of and changes to media filesXML file:

a simple format similar to SMIL, following commonly used internationalstandards and customized to effectively describe a typical lecture.

contains metadata fields using subsets of Dublin Core and IEEE LearningObject Metadata, describing media data

timing and synchronization information

Future Directions Allow more media streams (animations, data displays, chalkboards) Allow annotations, notes, links to other resources Transformations to more viewing objects: Quicktime, Media Player, iPod, etc.


Lecture Object - Example

Example of Lecture Object using current standard


Robotic Camera Tracking

Very easily distracted by anyreflective objects in the room

Inexpensive (4000 USD),simple and very accurate

“passive” IRUsed in Boeing factories

Requires extensive set-upand calibration

Almost reasonably priced(10,000 USD), more accuratethan RF

Ultra Wide Band (UWB)UbiSense

Slow and insufficientaccuracy, especially indoors

Almost reasonably priced(6,000 USD)

Radio Frequency (RF)BlueSoft

Requires extensivepermanent installation

Probably accurate enough fortracking a lecturer

UltrasonicDeveloped by AT&T labs

Very expensive, gets tiredafter several hours

Intelligent tracking thatgenerates pleasing video

Human camera operator

System Requirements Portable Robust Affordable Runs without expert intervention Little setup or calibration

Survey of Existing Technologies


Current solution: “active” IR system using dual-cameras and infrared necklace Target wears necklace made of IR LEDs Invisible to human eye, but very bright in IR Two cameras mounted together

• IR camera tracks target• visible-light camera provides color video

This system satisfies our criteria Portable: sits on a cart Robust: simple design makes it very robust Affordable: currently under 4,000 USD No expert intervention: start it and it just works Little setup: almost no calibration required Accurate to within centimeters

Robotic Camera Tracking – Current System


Remaining Challenges Rooms with windows: on bright days, excessive sunlight can increase

background noise and illuminate reflective objects Other IR sources

• Tablet PC’s use bright IR pens• incandescent light fixtures give off a lot of IR• transparency projectors give off a lot of IR

Improvements - Next system: machine-vision camera Stationary, wide-angle camera Frame rate, exposure time, timing are all controllable The necklace strobes

• identifies necklace uniquely to the machine vision camera• saves battery life

We are building and testing this system now This should greatly improve reliability in adverse conditions

Robotic Camera Tracking – Future Development


March 2005 Provisional patent filed, describing the current dual-camera with active IR

necklace.October 2005

University of Michigan researchers filed 287 invention disclosures in 2005 Eight of these were selected to be featured at the 2005 Tech Transfer

Inventor’s Reception, as the most commercially promising and world-changing.

Our tracking system was one of these eight.

March 2006 Permanent patent will be filed.

Technology Transfer


Plans for Using the WLCD Enable automated large-scale recording, streaming and archiving of

university courses Record and disseminate multiple parallel sessions at large professional

society meetings Facilitate the preparation of materials for distance learning Generate supplemental instructional content for outside-of-class review Provide services for recording of sample lectures given by faculty

candidates Recording of ATLAS Plenary Sessions Many more ...

Applications


Now - Physics 140 test Big impact: over 500 students each semester, including pre-medical,

engineers, science majors. Recording with current tracking system will begin March 2005 Two 1-hour lectures per week, each available through U-M CourseTools

within hours, and also podcast to student’s iPods. Chalkboard work will be captured. Entire system fits on one cart.

Fall 2006 – Pilot Project begins Project has support of the Provost and many academic units Three different classes will be recorded Student usage will be analyzed through surveys and log monitoring by

education experts in the Center for Research on Learning and Teaching(CRLT).

Project will continue through Summer 2006

Hot Item: U-M Campus-Wide Classroom Recording


Demo video made with tracking system in May 2005

Tracking camera video of Jen McCormick made in December 2005

Web Lecture for iPod - David Gerdes, UM Physics Professor

Example Videos