An Integrated Framework for Context-Enabled Discovery of Human Activity from Video

The Phase I effort of the SBIR program culminated in the development of algorithms, toolsets and interfaces to detect human activity from video and improve the operator"s wide area situational awareness of live feeds from multiple video sensors in the battlefield. In Phase II, we plan to extend this technology and achieve the following objectives: 1) utilization of scene context extraction algorithms, in conjunction with scene context exploitation algorithms for improved activity detection; 2) detection of suspicious events throughout the site of interest (regardless of if they are visible in the same camera view); 3) target-centric queries as opposed to area-centric queries (tripwire, area of interest breach) as well as several other UI enhancements; 4) view and camera invariant target matching for non-overlapping cross-camera target re-acquisition and tracking; 5) warning of threats before they actually materialize; 6) automated best shot analysis of the detected targets in the scene; 7) integration with distributed data sharing environments such as GHub; and 8) improved human activity discovery using novel features. In addition, metrics for activity detection evaluation will be investigated and a thorough performance analysis of the proposed innovations will be carried out.