Feasibility of Nonlinear Dimensionality Reduction for the Rapid Analysis of Persistent Surveillance Data, including the Detection of IED Placement Activity

摘要

Our ability to analyze large, complex data sets, such as persistent surveillance data, has often far outstripped our ability to rapidly analyze that data. We have identified a class of intelligent data reduction algorithms, known collectively as Nonlinear Dimensionality Reduction (NLDR), and we believe the utilization of NLDR approaches will allow a significant performance improvement for automated data analysis systems. In this report, we review the basic elements of NLDR techniques, we discuss the advantages of these techniques over more traditional approaches such as Principal-Component Analysis (PCA), and we outline an approach for utilizing NLDR to detect activities leading to the placement of IEDs based on airborne persistent surveillance video data.