DIPSARS: Discovery of Interesting Patterns and Semantic Analysis in Remote Space

The volume of data that NASA collects has grown dramatically as new missions come on line with hundreds of terabytes of data of all types. Often, this involves semi-autonomous exploration to collect a representative set of different scenes and objects so that research scientists and mission controllers can decide what regions are of particular interest and should be explored further. The lack of bandwidth to transmit all this data back from deep space has prompted the need to determine what data is relevant, important, and interesting enough to be sent back immediately. A common understanding among potential consumers (including the general public) of what signifies as an image or sensor reading of value is needed. To make these decisions quickly, the data needs to be presented intuitively using a common semantic vocabulary that describes interesting or anomalous features or events (e.g., composition of a certain mineral above some threshold, visual evidence of terrain features created by moving water, or a volcanic eruption in real-time during an orbital pass) so that opportunities for further inspection and analysis are not lost to long transmission times. We propose a framework for onboard intelligent data understanding algorithms to mitigate this opportunity risk and improve collection efficiency and effectiveness by summarizing the data and enabling the download of a subset containing the most valuable portions called Discovery of Interesting Patterns and Semantic Analysis in Remote Space (DIPSARS). DIPSARS is a framework for semantically characterizing, detecting, and fusing interesting and anomalous 1D sensor measurements and 2D imagery onboard remote spacecraft and planetary rovers to conserve transmission bandwidth. We do this by summarizing data events so that mission controllers and research scientists can analyze representative feature data and provide instructive feedback to seize emergent opportunities for data collection.

Advanced pattern recognition, knowledge modeling, and visual analytics are core business areas for Charles River Analytics, making the success of this effort fall squarely within our corporate interests and competencies. We will leverage our expertise in these areas to ensure the success of the technologies developed under the DIPSARS program beyond the end of the SBIR contract. The mature research developed under the DIPSARS effort will provide NASA with the capability to detect and semantically characterize interesting and anomalous 1D sensor measurements and 2D imagery onboard remote spacecraft and planetary rovers. Use of the capability will conserve transmission bandwidth by summarizing data using representative samples and semantic labeling so that mission controllers and research scientists can analyze representative feature data and provide instructive feedback to seize emergent opportunities for data collection. Primary NASA commercial application candidates include the XSearch and Monitoring Inductive Software System (AMISS) projects in support of potential NASA planetary rover missions (e.g., Mars Sample Return (MSR)) and celestial body observation missions (e.g., Mars Atmosphere and Volatile EvolutioN (MAVEN)). Other applications cited in NASA's Technology Area Strategic Roadmap for Modeling, Simulation, Information Technology and Processing may also become candidates with the maturation of the technology.

We expect the full-scope DIPSARS to have immediate and tangible benefits for a number of military planning and visualization systems. The capability will enable more effective semi-autonomous mobile robotic missions in the Special Operations and ISR communities by increasing the efficiency of operators and analysts during tactical situations. We also envision a number of commercial applications, particularly in undersea, underground, and polar region exploration. Urban search and rescue, and airborne weather pattern monitoring are other potential cross-marketing opportunities.We also plant to incorporate new DIPSARS technology into our VisionKit™ software tool to both increase its appeal as a commercial product and enable us to better provide consulting services to our existing customers. VisionKit™ (www.visionkit.com) is a suite of software libraries that reduces the cost and accelerates the development of real-time computer vision applications, such as those for self-navigating robots, security screening, and custom video processing. We will enhance VisionKit with Complex Scene Perception and Same Category but Different Type Detection capabilities developed under the DIPSARS project.