Spatio-Temporal Nonlinear Filtering With Applications to Information Assurance and Counter Terrorism.

摘要

This final technical report outlines research and main results obtained during the period from May 1 2006 through October 31 2011 of the MURI project. The objective was to develop a general and systematic foundation and algorithms for spatiotemporal statistical inference and for fusion of heterogeneous information from multi-source, multi-sensor distributed sensor networks. Immediate applications of the proposed work are Network Centric Warfare, where new and emerging systems such as MASINT and FORCENet collect but do not adequately interpret vast amounts of data; information assurance and network security; and homeland security applications, including video monitoring, and near-field and far-field intelligence analysis. Our research was targeted to solving three central problems: (a) nonstationarity, (b) integrating metric and symbolic information, and (c) very high dimensionality. Current methods for pattern recognition in monitoring and surveillance are designed for stationary patterns, and cannot cope with new patterns in ever- changing environments. We developed new statistical methods for the nonstationary environment, particularly spatiotemporal nonlinear filtering, changepoint detection, and advanced fusion methods. A distinctive feature of our approach is that the spaces in which estimation, classification and tracking is performed are both metric and symbolic.