Integrated Reconstruction and Labeling
R4-B4

Integrated use of physical models, machine learning, and multispectral sensing for direct material identification

  • Learn multispectral appearance model
  • Explicit metal artifact class + boundary model
  • Downweight less reliable data near metal
  • Directly estimate object labels and boundaries using graph-cuts

Relevance to the Homeland Security Enterprise

  • Mitigate image artifacts
  • Reduce number of corner cases
  • Reduce false alarms and need for OSARP
  • Reduce need for manual inspection
Project Leader
  • W. Clem Karl
    Professor
    Boston University
    Email

  • David Castañón
    Professor
    Boston University
    Email

Students Currently Involved in Project
  • Ahmet Tuysuzoglu
    Boston University