Is the security industry ready for autonomous AI?
The proliferation of intelligent video analytics powered by underlying artificial intelligence (AI) technologies has been among the biggest trends in the security industry over the past several years. However, skeptics of these solutions are quick to point out that true AI – computers with the ability to analyze a scene and make decisions own their own without a “human-in-the-loop” – is absent from these products and that the machine and Deep Learning algorithms they leverage are only as good as the datasets they are trained on.
But just as AI has improved across segments of the ever-growing Internet of Things (IoT) ecosystem, the same is also happening within security. Enter Corsight AI, which is looking to bring autonomous AI solutions to the industry for facial recognition applications. Borne out of Cortica, a developer of autonomous AI that simulates the neural processes of the human brain, the capabilities of Corsight’s technology extend well beyond what is possible with today’s Deep Learning algorithms.
“By basically trying to mimic the neural network of the brain, we manage to get programming as well as a data structure that give you speed and accuracy but also the ability to learn if an object is actually a different size or shape and get better results,” says Gadi Piran, the company’s CEO. “That is really different from Deep Learning.”
According to Piran, many of today’s popular Deep Learning solutions can be easily fooled if they encounter an object or a person in an abnormal situation.
“Imagine that you give a system a picture of a bottle. You put that bottle standing up, you take a picture and now a system with Deep Learning will know that if they can match the shape, it will be recognized as a bottle,” Piran explains. “But if a person in a video is holding that bottle upside down, Deep Learning usually has a very hard time saying, ‘this is a bottle.’ With autonomous AI, the concept is more of a true understanding of the object and then, no matter what position you give it, it will understand that it is still a bottle.”
One of the things that appealed to Piran about the technology developed by Cortica is that because one of their first applications was autonomous driving, they had the ability to simultaneously detect a multitude of objects at the same time, which stands in stark contrast to many of the video analytics solutions of old that had difficulties accurately identifying relatively low numbers of very basic, everyday items.
“Everything in the analysis, when it looks at the video in the situation, it literally detects everything in a microsecond and they have gotten to the point where they can distinguish between very intricate shapes,” he adds. “When I talked to them about facial (recognition) and so on, after a year-and-a-half discussion, we started in 2019 to do some development in the area and here we are in the beginning of 2020 starting a company based on the same technology but focused on facial recognition.”
Experience Matters
Having spent nearly 20 years on the VMS side of the market, Piran, the co-founder of OnSSI, which was sold to Qognify in 2018, says one of the biggest things that companies in the industry have always struggled with is defining the end result a product should achieve.
“Evolving from CCTV into what is now IP video surveillance, we saw the challenges building the Ocularis (VMS) product for a lot of companies. You get video, which is great, but the trick is to isolate events, meaningful events that are going to present themselves today,” he adds. “One of the challenges – because you’re always looking at people – was not only to detect an intrusion but also the identity of a person.”
In addition to delivering a reliable product for end users, Piran says they also want to ensure that Corsight is something that security integrators can install without a lot of headaches. In fact, the company has already completed a phase one integration with the Milestone and Genetec VMS platforms and they are also working on integrations with other video management solutions.
Addressing Adoption Challenges
Two of the biggest hurdles that Piran believes stand in the way to greater mass market adoption of facial recognition today are the aforementioned technical integrations with existing video systems and what he refers to as “deep integrations” with video data.
“It is connecting both ways; not only with video analysis and issuing an alert ourselves but also taking those alerts and moving them into the VMS so that it can process it,” he adds. “Those are deeper integrations and it is something that is going to take time.”
Perhaps one of the biggest stumbling blocks to the adoption of facial recognition are the privacy concerns that inevitably follow the deployment of such products. However, according to Piran, Corsight adheres to the strictest privacy standards in the industry, such as GDPR in Europe and CCPA in the U.S.
“Different countries and different regions have their own privacy regulations and one of the important pieces we put in the product was a) we do not record,” he says. “If you don’t want to, no visual images of persons are recorded, just the signature basically. The challenge for us, as a company, is to educate the video market, talk to the users and publish the guidelines on how to do it.”
The company is currently performing beta tests across several different vertical markets with its technology and Piran says that are also working to adapt it to address some the needs brought about by the COVID-19 pandemic.
“We have been asked by a couple of (government) agencies if we can detect people wearing masks. It actually was relatively easy and within a couple days, we came up with the capability to do recognition while people are wearing makes,” Piran says.
Joel Griffin is the Editor-in-Chief of SecurityInfoWatch.com and a veteran security journalist. You can reach him at [email protected].
