Artificial Intelligence (AI) and the promise held by Machine Learning (ML) and Deep Learning (DL) algorithms as it relates to video analytics has been one of the most ballyhooed trends in the security industry in recent memory. Unlike the intelligent video analysis solutions of a decade ago that overpromised and under-delivered in many respects, there is a belief by many in the industry that AI is truly going to be a game-changer in how end-users leverage surveillance systems moving forward.
Not only are these solutions expected to enhance security operations but it is anticipated that they will help organizations leverage video data in ways they never thought possible. And with the increasing number of IP cameras being installed in the U.S. and around the globe every year, the ability for humans to garner actionable intelligence from these cameras is simply impossible without the assistance of machines.
Though there have been a number of companies, both household names and upstarts, which have rolled out their own AI-powered video analytic solutions in recent years, Brent Boekestein, CEO of AI video analytics developer Vintra, says one thing many have failed to take into account is the impact that mobile video is going to have on the industry moving forward.
“Cameras are going onto things they typically haven’t (previously) been from a mobile surveillance standpoint. One of our thesis, which is kind of playing out in the market, is we’re not talking to any major enterprises or educational institutions right now that don’t have mobile as part of their next one, two, three or four-year strategy,” Boekestein explains. “What that means from an analytics perspective, all of your traditional (analytics solutions providers) are saying, ‘hey, we use AI,” and they do for classification but none of their tech works on mobile video. This is a big change in the market that creates a new opening for an entrant to develop a solution that works on fixed and mobile video.”
Founded in 2016, Vintra, which has a headquarters in San Jose, Calif., and a research facility in Barcelona, Spain, is debuting its FulcrumAI solution at ISC West 2019 in Las Vegas this week. FulcrumAI, which won the 2019 SIA New Product Showcase Award for video analytics, can take video from any kind of camera – fixed or mobile and from any manufacturer – to provide actionable data in any environment.
FulcrumAI can also run on-premises or in the cloud and features person, object and weapons detection along with facial recognition algorithms built from the ground up. According to Boekestein, there is a big difference between being able to offer true “detection” of things versus their “classification,” which is what he says is being offered by most video analysis solutions on the market today.
“Classification is I found something and now I want to know what it is. Most companies use something called background subtraction, so you cannot have a mobile video or use it on something like a PTZ camera, a drone, dashcam or body camera because the background is moving and therefore they can’t detect what is in the background and foreground,” Boekestein explains. “We’re using Deep Learning to both detect the object – whether you give me a dash camera, a drone, a PTZ or a fixed camera –and I can use the same algorithm to also classify it.”
Another thing that Boekestein says sets Vintra apart in the marketplace is that they are using their own proprietary Deep Learning technology whereas many other companies are using academically-available algorithms that they are embedding into their offerings.
“Think about this like going to Target, you buy it off the shelf or it is open-sourced on GitHub, you try to retrain it and you put it in your system and say, ‘We’re now an AI company,” he says. “I’m a baseball fan and that’s like me buying a Clayton Kershaw jersey and saying I’m a Cy Young winner. We built our own tech from the ground up. We’re not recycling someone else’s academically-developed algorithm.”
Boekestein says this enables them to run about two times faster than most of the other solutions on the market that leverage the algorithms developed by academia.
Moving forward, Boekestein says while starting with pre-built algorithms for things like object detection is a good thing, he believes the future of AI will be in allowing end-users to build their own models for things they want to track or detect which is something the company is currently working on to deliver to the market. “We call that the Analytics Foundry and we’re already working with one customer on a paying basis to develop custom algorithms for them costing many, many, many thousands of dollars to start – it will get cheaper over time – but the problems we’re solving are many, many, many millions for them,” he says.
Vintra sees educational and corporate campuses as well as the smart and safe cities markets as one of the sweet spots for their technology along with large transit system operators. Boekestein says they are also keenly aware of the tarnished reputation of video analytic solutions of the past and that they want to build trust with their partners and customers about the accuracy of their product and what it can accomplish.
“I think customers should definitely peel the onion back and do some due diligence on their vendors around how they’ve developed their products, what training data they have used, their accuracy scores on publicly available dataset, and so forth,” Boekestein adds. “That’s a big misconception; that just because I put ‘AI’ on it means its smart and I know what it is doing.”
About the Author:
Joel Griffin is the Editor of SecurityInfoWatch.com and a veteran security journalist. You can reach him at [email protected].
