'Smart' video offers an alert to threats
Taking boredom factor out of security systems
In video surveillance systems, the weakest link is the often bored, distracted human who has to spend hours staring at a bank of video monitors, waiting for something suspicious to happen.
Several Boston area companies say they have found a solution: surveillance systems smart enough to recognize threats, even when their human operators do not.
"It essentially replaces the need for people to watch video," said Scott Schnell, chief executive of VideoIQ Inc., a Bedford firm that was spun off earlier this year from General Electric Co.
VideoIQ is one of several local companies vying for dominance of the market for "video analytics" systems, which can help defeat a variety of security threats, from shoplifters to suicide bombers.
Systems from VideoIQ and Intuvision Inc. of Woburn can automatically spot an intruder climbing a fence or a subway passenger leaving a suspicious parcel on the platform. IntelliVid Inc. of Cambridge makes systems for retailers that can detect a possible shoplifter and electronically track his movement through the store, to see if he swipes anything else.
"It's really started to take off in the last two to three years," said Simon Harris, research director for security markets at IMS Research Ltd., a British technology research firm. Harris said that worldwide sales of smart video surveillance systems will be less than $100 million this year, but rise to about $3 billion by 2010.
Already there are more than 26 million surveillance cameras in the United States alone, according to a 2006 report from IBM Corp.
The research firm iSuppli Corp. said that worldwide sales of surveillance systems will climb from 30 million units in 2006 to 66 million in 2011. But these are mostly systems that require constant human supervision. There are not enough workers to watch so many million video streams. Even if there were, people make mistakes, perhaps fatal ones.
Computers that can recognize suspicious activities can serve as valuable backups to the human eyeball. The smart video systems are programmed to recognize certain shapes, like cars, people, or luggage. They're also designed to learn as they go, in the same fashion software can improve its recognition of human speech. The longer it runs, the better it gets at spotting anomalies.
In addition, the systems allow security managers to draw digital boundaries at key points. They can tell the software to ignore people walking past a building, but issue an alert when someone walks toward the front door.
The systems identify suspicious objects or movements by surrounding them with a box. For instance, the VideoIQ system puts a yellow box around moving objects that seem problematic, then uses its programming and experience to decide if it is a threat.
One test video shows ducks and boats on the Hudson River. The system draws yellow boxes around the harmless ducks, but when a boat appears, the box turns bright red. The computer does not dial 911; instead it sends an electronic alert to the video monitoring center, where a handful of people may be tending hundreds of monitors. The smart video system directs them to the suspicious video feed and can replay the event that set off the alarm. It's up to the humans to decide what to do about it.
Sadiye Guler, chief executive of Intuvision Inc., a video analytics firm in Woburn, admitted that computers still lag far behind humans at recognizing images. "It's still not as good as a 3-year-old child," she said. That's one reason humans are still needed to make the final call on a suspicious image.
Intuvision, a startup funded by grants from the US intelligence community, has attacked the problem using a technique called "task-based attention." This method teaches the machine to give maximum attention to certain events and much less attention to others. Just as a speech recognition program listens to samples of a person's voice to improve its accuracy, the Intuvision system views many images of people and objects and learns to identify them based on shape and size criteria.
A camera guarding a subway platform pays little attention to backpacks and suitcases that are moving. But it locks in on a package that has remained stationary for more than a minute or two, a package that might have been left behind by a would-be bomber.
"It does a shape analysis," Guler said. "Bag and person sizes are different. The shapes are different."
Intuvision's product is not yet on the market, but the company is deploying its first pilot system in the Washington, D.C., area. Guler won't identify the user. "I can tell that it is a federal law enforcement agency," she said.
While VideoIQ and Intuvision try to tap the homeland security market, IntelliVid Corp. of Cambridge has a different focus.
"We're really the only video-analytics-based company that specializes exclusively in retail," said IntelliVid chief executive Patrick Sobalvarro.
It's a massive market opportunity, given the scale of the retail-theft problem. "There's about $37 billion of this stuff leaving stores every year," with about half being stolen by store employees, Sobalvarro said.
IntelliVid technology isn't good enough to notice thefts of individual items. But if a sizable number of objects - say, razor blades - disappears from a shelf, the analytic software makes a note of it. The company's patented customer-tracking technology can see which visitor to the store removed so many razors. It can then track that shopper through every aisle, handing him off from one security camera to the next and looking to see what else he picks up.
Of course, the system also notes whether the customer heads for the checkout counter or straight to the door. If he tries to leave without paying, the systems sends an electronic alert to security workers.
A basic IntelliVid system for a small retailer costs about $10,000. It gets more complex and costly for bigger stores, which can have as many as 250 surveillance cameras per location. Sobalvarro won't identify his customers, but said that about a dozen of the nation's biggest retailers use IntelliVid systems.
Today's video analytics systems are designed to work with the millions of video cameras already deployed. But a new generation of smart cameras will make the systems more powerful and versatile.
"The trend really is to put the intelligence in the camera," by adding a built-in digital signal processor chip to analyze the video, said Harris of IMS Research.
Such systems could be cheaper than today's video analytics gear, because they won't need a separate server computer to collect and process the images. Instead, each camera in the network can be programmed to guard against different threats, because each will have a mind of its own.
Hiawatha Bray can be reached at bray@globe.com. 