'It's like Google for video'

The integration of facial biometrics and surveillance comes of age
Tuesday, May 1, 2007

Facial biometric technology used in access control is nothing new, but the use of facial recognition for surveillance purposes is really just starting to spark an interest among manufacturers, integrators and even some forward-thinking customers. Used notably in the banking industry today, integrators say it's a technology that customers--while they may not be asking for it by name--can readily understand and see the value in.
"It allows video to become a true investigative tool, both forensically and in real time," said Daniel Weiss, president and chief executive officer of Infrastruct Security, an integrator in Houston. "I think it's one of the most important trends I've seen in the last 15 to 20 years."
Many manufacturers are taking a good hard look at this technology, but the space is relatively uncrowded right now. 3VR and Dragnet Solutions are two companies that are active in this market, though they differ in their approaches: Dragnet Solutions sells a subscription-based service, while 3VR sells a DVR platform that is integrated with facial surveillance technology.
Gary Root, president and chief executive officer of Novato, Calif.-based Dragnet Solutions, said Dragnet's business model "revolves around providing a service to the industry. We don't sell products ... We build our platform to deliver that [surveillance] service to the industry." Root said, "the ability and knowledge and level of sophistication around operating a really massive biometric platform at scale is the essence of what our capabilities are." Right now Dragnet is focused on the banking industry, although Root foresees it fitting in with "broader initiatives" in the future.
Tim Ross, co-founder and executive vice president of 3VR of San Francisco, described 3VR's product as a "facial surveillance integrated into a full-service video management platform," and noted that "facial surveillance is simply one highly useful application that sits on top of our platform ... your transaction data, your case management system and video management system are all integrated into one product." 3VR does business in government, critical infrastructure, water treatment and petrochemical industries, casinos and retail.
So how does facial surveillance work? In banking, where check fraud is a $5 billion per year problem, a face is identified as a suspected fraudster, perhaps from a law enforcement database or because of suspicious behavior observed in one or more bank branches, and that face is put on a watchlist. The next time that face is recognized by a bank branch camera, it sets off an alert enabling security to intercept a fraudulent transaction and arrest the fraudster.
Ross said early attempts to apply facial biometric technology to surveillance camera feeds failed because the technology utilized straight-on, well-lit facial images designed for access control or passport applications. In a surveillance context, he said, it's difficult to get that perfect mug shot.
The key to good facial images is high frame rates, Ross said. "You don't need to store the video at high frame rates, you just need to be looking at high frame rates to find that moment" where the camera catches the best image that can be used for recognition.
Ross said 3VR licenses some biometric algorithms, but mostly uses proprietary facial biometric algorithms. Dragnet uses algorithms from Cognitec as a "template creation engine. It transforms the visual imagery into a searchable template," Root said. Dragnet then builds technology, such as video analytics and massive similarity matrix management, before Cognitec and after to make the platform actually work, Root explained.
At ISC West in March, 3VR announced that it had acquired Amcrin Corporation, developers of CrimDex, a database of thousands of known criminals used by close to 300 banks, law enforcement agencies, federal agencies and some large retailers. 3VR uses that database to augment internal watchlists. Ross described this acquisition as the "third leg of the stool" for his company, behind adapting facial biometric algorithms to a surveillance context and providing an integrated product to make a significant dent in the fraud problem.
Brad Stephanson, vice president of Diebold's physical security group, said in the future there may be shared databases of suspects which could be a service that an integrator could provide. "We are looking at that," he said, "and I assume others are as well. It could very well be that a 3VR or Dragnet does it and that we become a deployer and network consultant and that could be another scenario."
How do customers react to this technology? They realize it's the solution they've been searching for without knowing it, said Weiss of Infrastruct, which just completed a new demo room where it shows people 3VR's technology. "Once they see it, they get it," Weiss said.
Clients get frustrated, Weiss said, because they have "fancy video systems that the two times a year they want to use it, they can't find the image they want ... When you show them the magic of this thing, they say, 'That's Google for video.'"
Because it costs more than standard DVR technology, it's important to justify the premium to customers, Weiss said. "Anyone who's struggled with a time-lapse recorder and switcher or a time-lapse recorder and multiplexer and sat for four hours looking for that one slip-and-fall ... they see this and they go, 'Aaah.' There's the premium," he explained.
Customers can be won over when they see the technology, but for any manufacturer who's dealing with video analytics of any sort, Weiss said, they've got "a lot of work ahead of them to evangelize the channel ... it has to begin with the education of integrators' sales people ... I honestly believe that manufacturers forget that all the time," Weiss said.
What will the technology look like in the future? 3VR's Ross expects it to "recognize more attributes than just a person's face, by what they're wearing even in harsh lighting conditions or angles."
Stephanson said that Diebold expects facial recognition to be "tied with other forms for a multi-factor identification. It might be license plates; it might be voice. Coupled with this technology, you can increase the odds of a match."
When will it be mainstream? Stephanson predicted that if trials that are underway at several customer sites are successful, "we believe there will be large implementations underway in two years."
"Ultimately, it's the ability of the technology to effectively reduce costs that will drive the timetable for going mainstream."