Deep Architecture for Robust OPIR Tracking

Period of Performance: 01/01/2015 - 12/31/2015


Phase 1 SBIR

Recipient Firm

28 Corporate Drive Array
Clifton Park, NY 12065
Principal Investigator


ABSTRACT:Automated detection and tracking of moving objects in OPIR data is a difficult problem due to low GSD, low contrast, sensor artifacts, and other challenges, and many standard approaches will not produce sufficient result quality. A robust tracking solution, however, is highly desirable to aid analysts in the automated exploitation of OPIR data. Our proposed approach leverages deep learning to build a highly effective spatial-temporal feature set to support detection and tracking. Furthermore, we will incorporate multiple sources of information, such as GIS data, along with the pixel data to produce contextual layers. The purpose of these layers is to identify areas of the scene, such as clouds or oceans, that can aid or hinder the detection, tracking, and classification of moving vehicles. We will leverage Kitwares existing government open source Wide Area Motion Imagery (WAMI) software infrastructure, which supports real time processing and already has components integrated into GATER, as a baseline implementation and a framework for our proposed approach. Kitware will continue to deliver software developed on the effort with unlimited rights to the government.BENEFIT:Target detection and tracking in OPIR data is an important capability for analysts, such as those at the National Air and Space Intelligence Center or the National Geospatial Intelligence Agency. Manually detecting targets in OPIR data is difficult and time consuming, and significant human resources must be expended to build an intelligence product with manual tracks. The proposed technology will provide an automated tracking system for OPIR and related data that, when deployed, will dramatically reduce analyst workload and enable defense and military applications to be more timely and robust, resulting in much higher-quality intelligence. ???Additionally, the proposed technology has great commercial value beyond military intelligence, with potential applications in retail forecasting and logistics analysis; urban planning and environmental impact studies; insurance claim and fraud appraisal; and disaster response measurement. The technology will help address the lack of high-quality satellite video tools and provide commercial entities with a hardened, automatic system for video analysis and particularly mover tracking. This will result in significantly improved abilities to automatically produce saleable information from satellite video data.