HPC-MTT: High Performance Computing Enabled Multiple Target Tracking for Urban Surveillance Areas

ABSTRACT: Advances in large scale visual sensors enable (near) real time acquisition of imagery data over wide urban surveillance areas. Data collected by such sensors, such as the wide area motion imagery (WAMI) systems, however poses computational challenges to existing visual analysis algorithms. Among many tasks, tracking multiple moving target indicators (MTI) is of fundamental importance since it bridges the low-level imagery input (e.g., WAMI data) and the high-level situation awareness (e.g., trajectory prediction and understanding). In this project, we plan to design and implement a novel high-performance-computing (PHC) enabled multiple-target-tracking (MTT) framework, named HPC-MTT, with focus on large scale MTI tracking in the context of WAMI understanding. The design of HPC-MTT will be from three inter-correlated viewpoints. First, from the MTT point of view, we plan to parallelize all steps in the MTT pipeline, including background subtraction, MTI detection and multiple-target-association (MTA). Second, from the HPC point of view, we plan to design WAMI-specified parallelization algorithms following the partition-communication-agglomeration-mapping (PCAM) paradigm. In particular, we will design spatial PCAM, temporal PCAM and algorithmic PCAM to capture respectively the spatial, temporal and algorithmic characteristics of WAMI tasks. Finally, from the HPC platform point of view, we will deploy the proposed HPC-MTT in various HPC infrastructures including multiple-core computation (e.g., GPGPU for on-board tasks) and cloud computing (e.g., Amazon E2C-based prototype for off-board tasks). BENEFIT: The proposed high-performance-computing (HPC) enabled multiple-target-tracking (MTT) architecture has tremendous applications potential in many military applications. In C4ISR the proposed technology can also be incorporated into the systems that help warfighters process more information with greater efficiency, for example in the counter-insurgency and counter-IED area. In addition, some relevant Defense Acquisition Programs within DoD are such as Distributed Common Ground Station-Navy (DCGS-N), DCGS-X (Air Force), DCGS Army (DCGS-A) system, and AFRL E2AT and PCPADX. For some of these programs we already have close connections and know considerable program details. The market for military is quite large and IFT has successfully transitioned research prototype for inclusion in DoD contractor research simulations. We can also see commercial application of this project with the numerous concepts being developed. The proposed tool can be used to support commercial applications include medical diagnosis, quality inspection, disaster assessment, drug enforcement, air-traffic control, police departments, law-enforcement, search and rescue operations and anti-terrorist activity. Tracking algorithms that can solve complex ISR problems will have extensive commercial applications in a variety of areas such as timely emergency response and more comprehensive disaster monitoring. The size of this market is not small and may grow rapidly with the commercial demand in homeland security. We expect the aggregated market size will be similar to that of military applications.