NRL blends real-time video sources on SGI acquisition and exploitation supercomputer

By Bill MurrayGCN StaffThe Naval Research Laboratory helped develop software for a $75,000 Unix-based supercomputer to process full-motion, real-time video from different intelligence sources.NRL paid SGI $100,000 to help create its Video Acquisition and Exploitation System (VAES) over a two-month period, said Alan Dare, SGI's marketing manager for geospatial imaging.SGI officials delivered the first system on July 1, he said. When SGI gets more orders from NRL it will deliver more of the systems.VAES comprises an SGI Onyx2 Deskside supercomputer with customized software. It features two 400-MHz processors and runs the SGI Inix 6.5 operating system. It has 6G of RAM, a 1T hard drive and Fibre Channel RAID drives for storage, Dare said. The computer includes two digital-video-out boards for video acquisition, with 1,280- by 720-pixel progressively scanned video and 1,920- by 1,080-pixel interlaced video.'The whole idea is not to do something that is proprietary to the Defense Department, but to work with commercial partners' such as SGI and entertainment companies that want to use the technology, said Henry Dardy, chief scientist for advanced computing at the lab.'Our hope would be that operational sites would be interested in using' SGI VAES, particularly within the intelligence community, he said.The research lab's job is to 'keep pushing forward' by developing systems that DOD organizations can use, such as SGI VAES, Dardy said.SGI VAES can process video from handheld cameras, nautical vessels, surveillance aircraft and unmanned aerial vehicles.With that kind of video flowing through limited communications pipes, bandwidth is an important issue with SGI VAES, Dardy said.The system can handle 1.5G of data from a sensor'handheld camera, unmanned aerial vehicle, nautical vessel or surveillance aircraft'transmitted to the computer via satellite or other network, and can write the data to disk as well as display and process it at the same time, Dardy said.When used on deployment in an environment without a communications infrastructure, however, it could be difficult for U.S. forces to send video data to the supercomputer for real-time image processing because of bandwidth constraints, he said.'The link from the air to ground is the weak link,' in how SGI VAES operates, said Kirk Kern, an SGI systems engineer specialist. 'In developing the VAES, we focused on the camera capabilities and processing capabilities,' which are VAES strong points, he said.For much of the SGI VAES tests, NRL officials have used radio frequency or laser communications links to transfer the data from the sensor to the Onyx2 supercomputer. With NRL, SGI is working on using the MPEG image compression standard for unoccupied aerial vehicles to transfer video to the VAES, he said.NRL officials are using the same SGI Onyx2 supercomputer that SGI sells commercially with minimal hardware changes, Dardy said.For more than two years, NRL officials had been working on producing sensors that can process High Definition Television video, Kern said. They started working on processing the footage over the past six to eight months, he said. The National Reconnaissance Office provided some funding for developing cameras that work with SGI VAES, he said.Critical additions to SGI's commercial Onyx2 supercomputer include SGI Alias/Wave Maya Composer software for exploiting video streams, as well as a third-party developed software interface for I/O hardware, Dare said.The supercomputer uses an SGI high definition I/O board for input and Fiber Channel loops for output, Dare said.SGI VAES can take in television signals, including High Definition Television, and officials can ruggedize it for deployments. The Computer Ruggedization and Integration Division of RSI Inc. of Austin, Texas, helps NRL and SGI ruggedize the systems, Kern said.SGI VAES lets users import and edit video data in real time, with 'real time' defined as the ability to capture all video streams without dropping any frames, Dare said.The supercomputer can use up to four processors, but 'two are adequate for this job,' he said.The parallel processing capabilities of the deskside Onyx2 system are important, Dardy said. 'There's an architectural balance across the system.'Parallel processing allows users to perform at least two of the following functions at the same time: calculating background algorithms, saving video on the hard disk, downloading more data, and displaying, panning and zooming on video they're viewing on the Onyx2 supercomputer.'Higher resolution imagery is good. People like to see things happening as they're working with it,' including intelligence data, weather or research, Dardy said. They can also pan, zoom and rotate video to find local points of interest, he said.