Persistent Electro-Optical/Infra-Red (EO/IR) Wide-Area Sensor Exploitation

The ever-increasing size of persistent sensors requires automated tracking algorithms to be run on the aerial platform. However, transmission limitations restrict the automated analysis to forensic purposes to be off-loaded post-mission rather than providing actionable imagery that leads to timely intelligence. Our solution to this problem for gigapixel sensors is to transmit compressed image chips of moving vehicles detected by the on-board system. This solution provides actionable imagery within the limitations of the data link which is sufficient to allow tens of thousands of target image chips, compressed through zero-tree encoding of Haar wavelet coefficients followed by adaptive DPCM, to be transmitted every few seconds. Between appearance updates, the position and orientation of the vehicles will be updated on the ground station tracking display similar to pieces on a game board while reserving a portion of the bandwidth for full field of view updates every ten minutes. The minimization of false object detections is also imperative to reducing the demand on the data link when dealing with Massively Multiple Target Tracking (MMTT). We reduce false detections through context-based anomaly analysis, parallax mitigation, shadow compensation, and long-term information accumulated through the automated discovery of road networks. BENEFIT: The technology we propose to develop in this SBIR project will enable real-time exploitation of large-format imagery from ARGUS-IS and other gigapixel sensors. Currently it is impractical to extract much information from these sensors because of the sheer volume of data they produce. The proposed automated content extraction techniques will attract immediate interest from intelligence, mapping, and operational elements of the federal government.