Can this AI find a terrorist in a surveillance video?

By: Joe Gould 

Source Link

WASHINGTON — A Connecticut-based video analytics and cloud computing firm is pitching the U.S. Army on an artificial intelligence that can tag cars, people and weapons in drone footage, saving troops from drowning in data. IronYun used an appearance at the Association of the U.S. Army’s annual meeting to market a capability that uses multiple algorithms running simultaneously for intelligence, surveillance and reconnaissance — potentially saving thousands of man-hours of screen viewing. The idea of automatically searching through the reams of data brought in by military surveillance assets — often referred to as a “tsunami” by Pentagon experts — is an increasingly realistic thanks to emerging AI capabilities. And AI is an emerging priority for the Defense Department, which has several significant AI initiatives underway. Google has contracted with the Pentagon for similar AI-drone services, under the banner of the controversial Project Maven.

On Monday, President and CEO Paul Sun demonstrated at the company’s booth how, by using a series of drop-down menus, a user could search footage for a group of three people or more — or, hypothetically, a person with a blue backpack getting into a yellow car.

On display was a breadbox-sized appliance, called CityEyes, which can handle 32 camera feeds simultaneously, though according to Sun the system works across multiple platforms, with both analog and digital cameras. A separate server IronYun offers can handle hundreds of channels.

“We provide a software, just like we Google for text search,” Sun said. “It has face recognition, you can do a face search … It’s not just face, you can search for any sort of vehicle, from bicycle to motorcycle to trucks. You can do counting and you can identify vehicle colors.”

The system can also, almost in real time, alert a user when it sees that same yellow car, or any other object it is tasked to find, according to Sun.

The sophisticated facial recognition software can distinguish between different age and races, and as many as 30 facial features, like whether a person is bald, bearded or wearing glasses.

For the Army, the system could be trained to recognize its military assets — or an enemy’s, for intelligence purposes. “You can use a drone that can tell you there are 20 bad guys here and a cannon, or whatever,” Sun said.

Already the system has been deployed on drones in factory parking lots in Asia, to guard against theft. “Instead of having security guards constantly patrol, they have experimented with using UAVs to assist,” Sun said.

The software has been deployed in 15 different countries, in airports, cities and private companies. Government agency customers include the Drug Enforcement Agency and Federal Bureau of Investigation, according to Sun.

Asked about privacy concerns, Sun went to the example of a drug store client that uses the system. The system would recognize the face of a repeat small-time shoplifter, say, but not the person’s name.

“If I’m a troublemaker, I walk into the store, the store doesn’t want to know what the name is, but it’s a repeat offender the store manager knows,” Sun said. “The minute they walk in, the store manager gets a hit and they tailgate the guy to say, ‘You’re not our favorite customer.’”

Sun sees the system as useful to recognize a gun at a school to prevent school shootings. Or more mundane, to find a lost child in an airport.

“You could really save a mom a lot of grief,” he said.