Building Recognition by Function

Award Information
Agency:
Department of Defense
Amount:
$98,884.00
Program:
STTR
Contract:
W31P4Q-09-C-0051
Solitcitation Year:
2008
Solicitation Number:
2008.A
Branch:
Defense Advanced Research Projects Agency
Award Year:
2009
Phase:
Phase I
Agency Tracking Number:
08ST1-0178
Solicitation Topic Code:
ST081-009
Small Business Information
KITWARE
28 Corporate Drive, Clifton Park, NY, 12065
Hubzone Owned:
N
Woman Owned:
N
Socially and Economically Disadvantaged:
N
Duns:
N/A
Principal Investigator
 Anthony Hoogs
 Director of Computer Vision
 (518) 371-3971
 anthony.hoogs@kitware.com
Business Contact
 Anthony Hoogs
Title: Director of Computer Vision
Phone: (518) 371-3971
Email: anthony.hoogs@kitware.com
Research Institution
 UNIV. OF PENNSYLVANIA
 Mark West
 Moore Business Office
220 S 33rd St., room 280 Towne
Philadelphia, PA, 19104
 (215) 898-3150
 Nonprofit college or university
Abstract
The activity surrounding a building can provide functional information that is not available in static imagery. We propose to develop a method to classify buildings based on observed video activity that addresses the primary challenges of noisy, partial observables, and urban environments where buildings are close together. Our approach is based on probabilistic modeling of the activity surrounding a building, characterized by functional elements such as building entrances/exits, parking spots and traffic lanes. Activity patterns for each building type are learned over time from buildings with known type. Unknown buildings are classified by comparing their observed activity signatures against those for known building types. The method does not assume reliable tracking, does not rely on building geometry or appearance, and accounts for intermittent temporal coverage and buildings in close proximity. In Phase 1, we will determine which building types can be functionally recognized, and the associated requirements for video sensing such as resolution and extent of temporal coverage. We will also examine the effects of motion extraction accuracy on recognition.

* information listed above is at the time of submission.

Agency Micro-sites

US Flag An Official Website of the United States Government