DARPA awards additional $11 million for video search technology

Published 1 September 2010

As a result of advancements in intelligence gathering technologies (think UAVs), the U.S. military and intelligence community have been accumulating video archives over the past decade which make YouTube look puny; it is not only the number of pictures, but their quality: mere HD movies and TV are small and tightly compressed compared to the high resolution, full-motion imagery which pours in like an avalanche from every Predator or Reaper drone — and dozens of these surveillance drones are airborne above southwest Asia every minute of every day; DARPA is looking for an effective, automated video search technology

The Pentagon R&D arm DARPA has awarded $11 million to discover a method for a true video search. It is not only the Pentagon which is interested in video searching. Google, too, has been looking for ways to make video searching work as well as text searching does, among other things for the purpose of having ads show up alongside the ensuing video search results, thus making more money out of services such as YouTube.

Lewis Page notes, though, that the problems of finding the useful video out from among endless number of pictures of comical cats or doleful teenagers in their bedrooms, are nothing compared to the video search challenges the U.S. Defense Department faces.

As a result of advancements in intelligence gathering technologies (think UAVs), the U.S. military and intelligence community have been accumulating video archives over the past decade which make YouTube look puny. It is not only the number of pictures, but their quality. Mere HD movies and TV are small and tightly compressed compared to the high resolution, full-motion imagery which pours in like an avalanche from every Predator or Reaper drone — and dozens of these surveillance drones are airborne above southwest Asia every minute of every day.

Page writes that, often, all these mountains of super-quality video are seen just once, briefly, by the pilot and sensor operator handling a given drone via satellite communication from Nevada. “In future, as the machines become more automated, large amounts of perhaps useful video will never be seen by human eyes at all: there is simply too much to assign intelligence analysts to watch it all in the hope of catching something significant,” he writes.

Which brings us back to DARPA. The agency says that “Currently, video analysis for Predator and other aerial video surveillance platforms is very labor intensive, and limited to metadata queries, manual annotations, and ‘fastforward’ examination of clips.”

Translated into English, this means the U.S. military has generated — is generating — more video than it can meaningfully digest and analyze. DARPA’s solution is to look for a technology that can sift through mountains of video automatically. DARPA’s calls the project Video and Image Retrieval and Analysis Tool (VIRAT):

The software tools developed under VIRAT will radically improve the analysis of huge volumes of video data by: 1) alerting operators when specific events or activities occur at specific locations or over a range of locations and; 2) enabling fast, content-based searches of existing video archives.

VIRAT has been underway since 2008, and there have been previous awards: but on 27 August the Pentagon notified that New York firm Kitware Inc. had received a $10,962,069 add-on to a previous cost-plus deal — indicating that DARPA believes progress is being made.