Informedia VACE Quad Chart

1
Video Information Summarization and Testbed
VIDEO ANALYSIS AND CONTENT EXTRACTION (VACE)

• The Novel Ideas
• Building on the Informedia Digital Video Library
  infrastructure
• Fully automate the indexing of video streams
  based on multimodal information content
• Provide testbed environment for integration and
  demonstration of VACE Program developed
  algorithms techniques
• Extend from speech centric metadata extraction to
  indexing of image content and visual
  characterizations
• Video summaries and multimodal video mining
• Automatic creation of multimedia abstractions
• Automatic creation of video information collages
  as summaries over multiple events (e.g. timeline
  and geospatial tracking)

Informedia VACE System Overview
Information Collection Analysis
OFFLINE
Surveillance
Broadcast TV
UAV
Digital Encoding
Indexing
Analyst
Motion Analysis
Object Extraction
Speech, Face, TextRecognition
Relevant Result Set

• Milestones/Dates/Status
• Integration Platform Mo/Yr
• Installation of demonstrator system at DIA and
  NSA Dec 2000
• Publication of interface specifications and
  procedures Mar 2001for data and module
  integration
• Integration of two collaborators into
  demonstration system Aug 2001
• Modularization of processing architecture Apr
  2002
• XML/XSL data interface and display templates Aug
  2002
• Visualization/Summarization
• Geographic collages with zoom in/out
  visualization Mar 2001
• Interactive multimodal queries Aug 2001
• Time based collages as event summaries Apr 2002
• Link analysis of metadata co-occurrence Aug 2002

• Impact and Champions
• Impact
• Provide unified infrastructure for integration
  and demonstration of systems for
• Object detection, recognition and tracking
• Audio, speech and natural language
  understanding
• Automatically generated video information
  summaries and multimodal video mining
• Enable the evaluation, testing and validation of
  research results by analysts in user-focused
  application
• Champions
• DIA, NSA

Lead Principal Investigator Howard Wactlar,
Carnegie Mellon University
Project COTR William P. Olexy, DIA
10/00