A Real-Time Robust Autonomous Surface Threat Identification and Visual Knowledge Discovery System

A set of robust software tools for automated identification/classification/recognition and tracking of surface threat among a group of non-threat entities is vital in the development of a good surveillance system. The increase in volumes of imagery data imposes a paradigm shift in the current surveillance systems used primarily only after the fact as forensic tools, to automatic knowledge exploitation tools to discover the relevant knowledge, forecasting the onset of (possibly catastrophic) events prior to their occurrences - such knowledge being critical for preventive actions across a number of government agencies - to detect, track and monitor (suspicious) objects and humans, their patterns of behavior and their associations in order to prevent possible calamities. Currently this capability is not available due to inadequate feature extraction and intelligent analysis tools that can in time communicate critical information collected and analyzed by various sensors. The objective of this project is the feasibility of an innovative Visual-Knowledge-Discovery system (VKD) to robustly and rapidly analyze actionable knowledge from visual data to identify and classify potential threats among a group of non-threat entities. The design and development of the system is based on a hierarchical/multi-resolution scene decomposition scheme to localize, recognize and track movement of objects of interest.