A Scalable Architecture for Operational FMV Exploitation

A scalable open systems and standards derived software ecosystem is described for computer vision analytics (CVA) assisted exploitation of full motion video (FMV). The ecosystem, referred to as the Advanced Video Activity Analytics (AVAA), has two instantiations, one for size, weight, and power (SWAP) constrained conditions, and the other for large to massive cloud based configurations. The architecture is designed to meet operational analyst requirements to increase their productivity and accuracy for exploiting FMV using local cluster or scalable cloudbased computing resources. CVAs are encapsulated within a software plug-in architecture and FMV processing pipelines are constructed by combining these plug-ins to accomplish analytical tasks and manage provenance of processing history. An example pipeline for real-time motion detection and moving object characterization using the flux tensor approach is presented. An example video ingest experiment is described. Quantitative and qualitative methods for human factors engineering (HFE) assessment to evaluate cognitive loads for alternative work flow design choices are discussed. This HFE process is used for validating that an AVAA system instantiation with candidate workflow pipelines meets CVA assisted FMV exploitation operational goals for specific analyst workflows. AVAA offers a new framework for video understanding at scale for large enterprise applications in the government and commercial sectors.