Abstract
Organizations are starting to realize of the combined power of data anddata-driven algorithmic models to gain insights, situational awareness, andadvance their mission. A common challenge to gaining insights is connectinginherently different datasets. These datasets (e.g. geocoded features, videostreams, raw text, social network data, etc.) per separate they provide verynarrow answers; however collectively they can provide new capabilities. In thiswork, we present a data fusion framework for accelerating solutions forProcessing, Exploitation, and Dissemination (PED). Our platform is a collectionof services that extract information from several data sources (per separate)by leveraging deep learning and other means of processing. This information isfused by a set of analytical engines that perform data correlations, searches,and other modeling operations to combine information from the disparate datasources. As a result, events of interest are detected, geolocated, logged, andpresented into a common operating picture. This common operating picture allowsthe user to visualize in real time all the data sources, per separate and theircollective cooperation. In addition, forensic activities have been implementedand made available through the framework. Users can review archived results andcompare them to the most recent snapshot of the operational environment. In ourfirst iteration we have focused on visual data (FMV, WAMI, CCTV/PTZ-Cameras,open source video, etc.) and AIS data streams (satellite and terrestrialsources). As a proof-of-concept, in our experiments we show how FMV detectionscan be combined with vessel tracking signals from AIS sources to confirmidentity, tip-and-cue aerial reconnaissance, and monitor vessel activity in anarea.