With recent advances in machine learning capabilities, a new arms race is underway among video surveillance software developers to determine who will be the leaders and laggards in this brave new world of artificial intelligence (AI)-powered analytics. From facial recognition to object and weapons detection, there are a whole host of new intelligent video analysis solutions making their way onto the market today, many of which are from companies that are relative upstarts in the security industry.
The latest company to throw its hat into the AI surveillance arena is Defendry, which specializes in providing automated threat detection technology to a variety of vertical markets. According to Pat Sullivan, Defendry’s founder and CEO, the company was initially focused on the development of a project and incident management software platform called “Ryver,” which helps workers within an organization collaborate and delegate tasks among themselves more efficiently. The company shifted gears earlier this year, however, after it acquired AI startup Deep Science, which was founded by Sean Huver, who now serves as Defendry’s Chief Science Officer.
“When I decided I wanted to do this and wanted to focus on this particular area, I looked at everybody who said they had a weapons AI dataset. I looked at several but I couldn’t find any that actually worked until I found Sean’s company. Over a period of time we got to know one another and decided to put the two companies together,” Sullivan explains.
According to Huver, Deep Science was borne out of work he did as a research scientist on projects for the Department of Defense in developing technology that could detect drones carrying explosives. He soon realized, however, that this technology could be converted into a solution to address more common, civilian-based threats.
“I had a local 7-Eleven near me, in a relatively safe neighborhood, robbed for the first time. The owner was pistol-whipped and knocked unconscious and when they went back and looked at the video surveillance footage, the guy who did it spent about 15 minutes kind of rummaging through the store, taking a bunch of lottery tickets and trying to open the cash register just before giving up and running off with the lottery tickets and cigarettes,” Huver says. “He knew that he had all the time in the world, that the cameras were, of course, passive devices and so that was the impetus that made me want to apply what I had learned on the DARPA project and really build this active threat detection solution.”
The combination of the companies’ products has enabled them to offer what they refer to as Active Response Technology or ART, for short, which automatically detects and sends alerts about threats, such as an active shooter or robbery, in real time. “The goal of the product and the vision of the product are to instantly recognize a weapon or mask and lock the doors automatically with the goal of preventing an active shooter situation or robbery from happening,” Sullivan says.
Unlike other AI solutions that merely alert relevant stakeholders when a potential incident is detected, Sullivan says that Defendry’s notifications are sent to a 24/7 monitoring center where a human verifies if it is real.
“The AI, when it sees something, it sends an alert with pictures and other information in front of a human – what’s called a human-in-the-loop – who is monitoring that particular incident and gets to make a very quick decision whether this is real or a false positive, which can happen,” Sullivan says. “Ryver is the underlying platform that makes the AI work because the AI can see something but if the best it can do is send a text message or an email to somebody that’s not very good and, with most systems, that’s what they do.”
Although Defendry’s product is ideal for a wide range of use cases, Sullivan says they have initially been in conversations with large retailers, such as convenience store and discount chains, about how the software could be used to mitigate robberies and burglaries within their establishments.
“They all have cameras but they know no one is watching. Literally no one is watching those cameras,” he says. “The can use those cameras to learn what happened but they have no way to respond to it actively in real time. Another market that we are finding huge interest from is places of worship – churches, synagogues, and mosques – because they all know they are targets.”
When it comes to helping increase the adoption of AI solutions in the security industry, Sullivan sees the primary challenges as being sales and marketing and scale.
“The technology works. It is a matter of finding the best way to get to the market,” he says. “Integrators are our way to get to the market. We have tons of conversations going on with systems integrators who have been selling cameras and access control but they have never sold artificial intelligence anything. They know it’s the future and they now know the future has arrived.”
Another barrier to adoption, according to Sullivan, is the notion that the use of AI will potentially introduce bias into how video surveillance is leveraged which he says couldn’t be further from the truth.
“One thing we tend to hear is concerns about facial recognition and are you profiling somehow or are we violating some law or at least some politically correct approach and our answer to that is no. We don’t care who is holding that gun and we’re not recognizing their face, we’re recognizing an object,” Sullivan adds. “We point out that all we’re doing is looking for people with guns and masks, which has nothing to do with their race, color, creed or whatever.”
About the Author:
Joel Griffin is the Editor of SecurityInfoWatch.com and a veteran security journalist. You can reach him at firstname.lastname@example.org.