A framework for fielding Artificial Intelligence techniques on High Performance Computing (HPC) Hardware

The ultimate goal of this proposed effort is to improve the ability to determine suspicious radar track behavior and suspicious areas to focus attention on. We propose a framework for representing the knowledge and human-quality reasoning required to process large quantities of radar data, transforming it into a form that can be efficiently executed in real time on an HPC system, and then efficiently executing it given the actual dynamic tactical situation. In addition to being scalable up to a large number of pixels and objects, it should also be adaptable both in the long term to different HPC configurations and in real time to different dynamic computational loads. In Phase I we will study the urban environment and resulting radar data; determine the tactical reasoning and available information that can be applied to knowledge-based frame-to-frame radar track correlation and vehicle behavior analysis; elaborate the heuristics and algorithms for learning traffic patterns, normalcy, and other knowledge from past data; develop techniques for the automatic translation to a form for HPC hardware; develop real-time HPC resource scheduling techniques; prove the feasibility through prototype development, experimental testing with real radar data, and demonstration; and develop the Phase II system design.