UAV Battlelab Experiments With Feature Recognition Software
By Jefferson Morris
04/13/2004 10:05:18 AM
The U.S. Air Force's Unmanned Aerial Vehicle Battlelab (UAVB) has tested software that can pick desired features out of UAV video long before they become visible to the naked eye, according to Lt. Col. Timothy Cook, chief of the UAVB's Combat Applications Division.
Based at Eglin Air Force Base, Fla., the UAVB's mandate is to take existing hardware and weapons and integrate them with UAVs. The recognition software originally was developed for the Nevada gaming industry, to automatically spot problem gamblers when they enter casinos, Cook said.
"It's based on the dimensions of your face," he said. "If I trained the camera on your face, it would plot the distance between the pupils of your eyes ... the length of your nose, the width of your mouth."
UAVB personnel were given a demonstration of the software during a trip to the MGM Grand Casino in Las Vegas, which has 1,600 cameras monitoring activity 24 hours a day. The software was told to look for the face of a particular person, who then walked around the casino floor past different cameras.
"Every camera he popped up on put a red circle around his face and started beeping," Cook said. "Everywhere he went, it could trace him." The software even identified a reflection of the man's face in a pane of glass, he said.
Intrigued by the possible applications to UAV surveillance video, the UAVB conducted a test last year at Eglin using streaming video from a Pointer UAV. A captain's face was entered into the computer as a search item, and the UAV was launched.
"It starts beeping on this clump of trees," Cook said. "And they had to drive the UAV about another two miles before they could get close enough [to see] there was a vehicle underneath the trees."
The captain whose face had been loaded into the computer was sitting in his truck eating lunch. "It found his face through the trees, through the windscreen, in the shadows of the trees, and we went, 'Wow, we need to explore this,'" Cook said.
Dubbed DIVOT (Digital Imagery and Video Object Tracking), the software later was put to work on pre-recorded video taken by a Predator UAV in Iraq. The system was provided with imagery of certain objects, then told to identify them in another video.
"The scene is a flat desert with some black clumps on it," Cook said. "And when the Predator is about 10 miles away, it starts beeping on one of the clumps. And it takes probably five minutes for the Predator to fly close enough where you can finally make out with the human eye that it's even [an object], let alone the one that we told it to find."
DIVOT can compress video at rates of up to 400 to one, then search through the compressed data very quickly, according to Cook. It also features 256-bit revolving encryption, which should make it more than up to the standards of the National Geospatial-Intelligence Agency (NGA), he said.
While the UAV Battlelab continues to pursue DIVOT's application to Predator operations, the technology also has found its way into the classified world, Cook said.
"Now this thing is kind of out of our realm right now," he said. "Because of certain customers and certain platforms it's gone back into the black."
Рецидивист под пристальным оком спецнадзора.
(для трижды уважаемой администрации: всё вышесказанное--сугубо моё личное скромное субьективное ХО, ни в коей мере не претендуещее на правду в последней инстанции, и основанное исключительно на моем индивидуальном восприятии.)