Credit union lauds facial search application

MECHANICSBURG, Pa.—Does facial recognition technology work? For Chip McBreen, vice president of fraud and security services at Members 1st Federal Credit Union it does. “It's allowed me to gather additional information that I might not know is out there,” he said. “In a strict date and time search, I'm looking for a specific event and that's what I'm getting. With the addition of facial recognition, I'm able to do an enterprise-wide search, allowing me to gather evidence that I just didn't know about.”

He has even, on more than one occasion, experienced the best-case scenario: A suspect from a watch list entered a branch, was picked up by the facial recognition, and McBreen's staff stalled the suspect until the police arrived and apprehended him. “How important is that? I'm a retired trooper and I can tell you it really improves your relationship with a police force when the victim—that's us—is actively looking for your suspect ... All of the departments just love us to death. When we give them evidence, it's all tied up in a bow. We give them a DVD, the images are crystal clear, and it has all the information they need embedded right on it.”

How did Members 1st get to this place?

Frustrated with a PC-based NVR system that was failing after about three years of operation, McBreen asked his integrator, Gilbertson Group, for a better solution. Matt Gilbertson, Gilbertson president, had just started selling 3VR-based systems, which employ facial- and license-plate-recognition technology, along with a suite of other analytics to make video search more efficient.

“I was a bit of a skeptic,” McBreen said.

But so was Gilbertson. “When the rep said he had a new kind of DVR to show me,” Gilbertson said, “I told him I needed a new DVR like a hole in the head.” Gilbertson, with 20 years in business in nearby Coatesville and 45 employees, pride themselves on their internal testing facilities, though, and after nine months of running the system at its 30,000-square-foot headquarters, the company was ready to start deploying it. “We wanted to make sure all of our guys are very well trained on the system,” Gilbertson said, “because even though it's a fairly easy end user interface, it requires an attention to detail on the set up and the programming side in order for it work well. It's a lot of system to know before it can just go out there; it's not something you just hook up cameras to and it works miraculously.” That said, now that the bugs are worked out, “We're selling a ton of it,” he said.

McBreen has 43 branches and said by 2011 every branch would be part of the 3VR system, which allows search of video enterprise-wide.

“Everything looks good in a demo,” McBreen said, “but when you put it out there in a live scenario, my experience is that it only works half of the time. So we agreed to test this. I purchased one and put it in our busiest branch. And I didn't break it. I was pleasantly surprised that all the bells and whistles actually worked, and I didn't have to change my whole strategy, or build my data room in each branch around this unit. It fit right into our existing plan.”

For the facial piece to work, though, both McBreen and Gilbertson emphasized a need for good facial recognition policy. Members 1st requires all customers to remove hats and sunglasses, for example, something McBreen had to fight for with a management group that emphasizes customer service. Further, Gilbertson designed specific chip-board-based megapixel cameras to sit on the walls behind tellers, along with domes gathering images at choke points set up at entrances and exits.

“With every specific location, there may be some necessity to tweak the cameras a bit,” Gilbertson said, “but what we do differently is set our systems up in our facility, run them through the paces, try the analytics, test out various cameras with various light conditions, to see how well they work or don't work, so that we can deploy them much more successfully. We had to relocate a few cameras when we first started with Members 1st. Now it's down to a science.”

Neither McBreen nor Gilbertson had numbers to provide for false-alarm or positive-capture rates, but Gilbertson said he is rarely asked for a false-positive number, and McBreen said it was sort of irrelevant to the way that 3VR delivers results.

“I can tell you that from an end user perspective, the false positives we get are negligible,” McBreen said. “The false positives are easily corrected with a click of the mouse, and quite honestly, I might get five images and two of them are my guy, and three of them are not, but they're close. If that happens, I just ignore those three. It did find the guy I wanted and to me that's a much bigger deal. Your brain just has a way of tuning that out: That's not him. There he is. There he is.”

“We don't get a ton of pushback on that,” Gilbertson agreed. “We don't try to oversell any of the analytics right off the bat, because everyone has a different need, and I think 3VR took a unique approach, based on a confidence level. We realize we're going to get some false positives, but with the event cards that come up as we're searching for video it's easy to see, 'nope, that's not the guy,' and we don't really care about the false positives. They'll be there, but there's an easy way to filter through how the system gives us the results.”

Gilbertson chuckled. “I think what really changed our approach was that when we were explaining what this product could do, we scared the hell out of people, because they were imagining a $50,000 recorder. But it's really just not that expensive.”