In recent years, a synchronicity of progress in cameras, lighting and processors – all stirred together with hundreds of thousands of man-hours of development – has brought machine vision to the edge of general deployment. Some futurists think that’s very big news. The only thing preventing robotics from having the revolutionary effects anticipated for so long, they say, has been their inability to see. (Technologist and author Marshall Brain believes that vision-enabled robots will disemploy more than a third of the U.S. labour force during the next 20 years.)
Whether that happens or not, machine vision is almost certainly going to revolutionize security. The technology never gets distracted, forgetful or tired. It is network-compatible, scalable, readily upgradable, searchable and archivable. Like all things digital, its price is on a one-way trip to zero. Machine vision not only touches nearly every responsibility of the security mission as currently defined, but it seems likely to rewrite both the meaning of security and its relation to the organization as a whole.
Perhaps the machine vision application now spreading fastest is what Lee J. Nelson, principal systems consultant for Electro-Optical Technologies Inc., calls intelligent optical character recognition, or IOCR. It differs from unintelligent OCR in that the former recognizes characters printed on physical objects – often moving, dirty and outdoors – as opposed to ones on a printed page.
According to Donald Brick, president of Hi-Tech Solutions USA, a machine vision company, IOCR is now being used at many ports around the world to keep track of containers. It can identify which containers have arrived, where they’re stacked and if they’ve been loaded onto a particular ship. The most important IOCR application, by sheer number of installations, is probably license plate recognition (LPR), currently used for objectives ranging from intrusion detection to toll collection.
In the context of the parking garage, the primary benefit of LPR is in controlling the lost ticket con. Every garage sets an amount that has to be paid by people who have lost their tickets. Anyone running a bill higher than that amount has an incentive to throw the ticket away. As time goes on, that incentive becomes more powerful. According to Louis Vinios, president of JPA Management, a building management company, the con is chronic and serious, since it always involves large amounts of money. And, while you could send an employee around to do manual entry of license numbers, the task bores people quickly (and bored humans tend to work slowly, expensively and inaccurately).
Vinios has no hard figures, but he suspects that the Trakker LPR system he installed in the Radisson Hotel he manages in Boston choked off enough bogus lost-ticket claims to pay for itself in a year. There are other benefits too. Vinios says he is sometimes in a position to help law enforcement (two of the 9/11 hijackers left their cars in an LPR-equipped garage), and occasionally customers ask for precise records of their comings and goings for an audit trail. Consultant Nelson adds that building security managers often have reasons independent of revenue for wanting to keep track of who’s parking in their buildings.
As important as IOCR is, the big payoff for machine vision in security will be in object and behaviour recognition. Consider change recognition, for example. You might want to compare the underside of a vehicle or a container using a reference image to look for differences. Changes might be evidence of tampering; if any were to be found, the vehicle in question might be flagged for special treatment, such as X-ray examination. Brick of Hi-Tech Solutions predicts that in the near future, owners of vehicle fleets, valet services and parking garages will also use such systems to recognize damage (false damage claims are another risk for these businesses).
Machine vision is far more flexible than the usual tools of physical security. “After 9/11, we installed cameras along the Mexican border,” says Bill Anthony, spokesman for the U.S. Bureau of Customs and Border Protection. Since there were far too many cameras for the number of agents available to monitor them, the bureau tried to filter the camera feeds with motion detection.
People recognition
Machine vision is far more flexible than the usual tools of physical security. Unlike cameras, people-recognizing systems can look for patterns of movement associated with humans.
That didn’t work. “There are a lot of rabbits out there,” Anthony says. “We had to check each alert, and by the end of the day we were getting tired.” So the bureau bought a people-recognizing system built by ObjectVideo. It’s a program that looks for patterns of movement that are associated with those made by humans, like a sphere (a head) nodding on top of a cylinder (a torso). According to Anthony, the upgrade, which was rolled out over the existing infrastructure, has eliminated the false positives. Recently, the bureau expanded its contract with ObjectVideo to cover all points of entry, including international locations.
Another application for people recognition is to detect piggybacking and tailgating. Piggybacking refers to cases where a person authorized to pass through a control point allows another to do so, perhaps because the latter is waving a pass or performing a routine maintenance activity, such as waxing the floor. Tailgating is when unauthorized people crowd in behind an authorized access.
Such tricks can be defeated with turnstiles, mantraps and guards, but those techniques have costs and limitations. “Mantraps and turnstiles are physically cumbersome. They’re hard to move traffic through, and they limit your access points,” says Jerry Brady, CTO of Guardent Inc., a security services company. “Machine vision would allow people to pass through any number of entrances and exits while gathering information on the traffic flow. I’d love to use it.”
He certainly could, as there are a number of piggybacking recognition systems on the market today. However, before Brady makes a major investment, he wants to see a system that offers across-the-board integration with enterprise IT and other security technologies, from biometrics to wearable panic pins.
Integration issues
The digital nature of machine vision raises integration issues. For instance, up to now it has been impossible to hook more than a very small number of videocamera outputs to a LAN, since a small number of video outputs can suck the bandwidth out of even a high-speed Ethernet. Run locally (as it always is), machine vision acts like a compression algorithm, putting only actionable data – exception reporting, in essence – with maybe a couple of illustrative .jpegs onto the Net. That makes it possible to integrate the outputs of hundreds of cameras around the enterprise.
One implication is that CSOs will be asked to define a new class of security issue: not violations, but flags. A car recognition system might count how often the same car parks in remote corners late at night, at least raising suspicion. Or it could count how often someone drives around the perimeter – two times around might only mean they are lost; four times is more interesting. A stranger opening one door inappropriately could be cut some slack, but if the same stranger were to open, say, three doors for no reason, even over several days at different locations, he might be questioned.
Second, an enterprise landscape view, inside and out, could be useful to other divisions and departments. For example, personnel might be interested to learn that certain employees are habitually working into the wee hours of the night; facilities management would like to know when some important piece of manufacturing equipment produces an unusual vibration that presages an impending failure; and customer support might draw useful insights from the fact that customers predictably cluster in certain areas of a store. A CSO sitting on such a powerful tool could find reason to organize an enterprise-wide steering committee to help accommodate and respond to these additional interests (and perhaps pay for the necessary infrastructure). Similarly, once a security force is freed from having to sit and watch banks of monitors, it might be able to help meet other important security needs.
Until now, most enterprises have viewed security as a function grafted on by necessity, but basically not a part of the company culture. Security has been seen as generally isolated from the enterprise mission and, at bottom, not shown much appreciation. Machine vision may change that. If security is the first department to roll out this very powerful and significant technology for its own purposes, it might well find itself asked to participate when the enterprise decides to leverage it for other ends.
