Building Recognition by Function

The goal of BLUE is to identify the function or purpose of buildings using persistant surveillance video such as Constant Hawk or Angel Fire. In Phase 1, we developed preliminary algorithms demonstrating that functional recognition of buildings and other scene elements such as parking spots and doorways is possible, using both supervised and unsupervised techniques to define functional models of building and scene element categories. On Phase 2 we plan to enhance the unsupervised method with elements of the supervised method; extend the set of track features; add comprehensive scene modeling to include relations between scene elements and buildlings; address outstanding research issues such as dealing with poor quality tracks; and test extensively on wide-area video. The resulting algorithms will be implemented in a prototype system integrated with our video processing framework, andsuitable for transitioning into operations.