Features Video Surveillance Week
The next phase of surveillance architecture

The evolution of artificial intelligence (AI) combined with video analytics is offering significant improvements when it comes to video surveillance.

“Video analytics provides a useful aid to manned surveillance. It can provide live alerts, for example, a person entering a perimeter; or analysis of recorded video – for example, search for a person wearing a red top,” says Jon Cropley, principal analyst, Novaira Insights.

“Apart from security applications, it can also aid with business intelligence. For example, counting the number of people entering a store or monitoring queue lengths.”

AI has played a significant role in the optimization of video analytics in recent years.

“The number of real-world applications that AI and analytics can feasibly solve is both rapidly expanding and providing needed improvements in accuracy,” says Jammy DeSousa, associate director, product manager, Johnson Controls.

But how is AI guiding these technologies and how much of this is occurring within the actual device versus the cloud?

“It depends on the supplier. The way that products are designed, it all depends on what you’re trying to achieve,” says Greg Tomasko, applications engineering leader for Honeywell Building Technologies (HBT).

“In my experience, there’s a lot of analytics that are being driven now out to the edge, and that was influenced by a lot of the camera manufacturing around the world just getting smarter and faster and more powerful in their chipsets.”

DeSousa says that in today’s market, to be a leading manufacturer you must offer an AI-enabled camera device in your portfolio.

“Today’s camera technology presents computing power that is easily suited for very powerful video AI. In the future, we can foresee hybrid architectures where cloud hybrid platforms could take data analysis to another level. However, for the time being, camera-based AI is in its golden moment,” he says.

One company that does run most of its AI in the cloud is cloud video surveillance company Eagle Eye Networks. Dean Drako, CEO of Eagle Eye, says this gives his company tremendous advantages.

“We transmit most of the video to the cloud all the time,” Drako says. “AI in the cloud is better, because it means that Eagle Eye is maintaining the servers, maintaining the models, keeping it up to date. We can upgrade it, keep it running, and it’s no headache for the customer.”

Cropley says a range of architectures are available to deliver video analytics.

“It is unlikely one single approach will win out entirely with different approaches for different scenarios. It is likely hybrid architectures including a combination of the edge, an on-premise server or NVR, [and] in the cloud will become a popular approach for the deployment of video analytics,” he says. “Considerations when designing the mix between these approaches include processor requirements in the camera, camera lifecycles and barrier to adoption to cloud.”

Cropley says his company estimates that almost 20 per cent of network video surveillance cameras featured edge hardware accelerated deep learning (AI on the camera) in 2022, and this percentage will grow quickly over the next five years.

Systems integration

But what does this mean for the systems integrator that installs the camera network and how does this affect their ongoing relationship with their client?

In the case of integrators installing HBP cameras, which have had many of the latest applications built into the units, Tomasko says they become a much more powerful partner.

“We manufacture these products, so it’s not my team that’s out there pulling wire and hanging cameras. We want to make sure it’s open and scalable and flexible enough to deploy in 90 to 99 per cent of the applications available,” Tomasko says.

“When the customer asks, ‘What about analytics? How can I get more intelligence?’ The response from the integrator doesn’t have to be, ‘Oh, we’re going to have to redesign, and we’re going to have to talk about bandwidth and processing.’ It’s, ‘That feature is in your camera. Why don’t you just turn it on?’ We can show you that value, and it makes [the integrator] a more powerful partner to their customers because they don’t have to have such a steep learning curve to jump over the analytics barrier.”

DeSousa adds that it demonstrates that the systems integrator is up to date on modern technology and practices.

“It also reinforces the trust in the relationship between the systems integrator and end user,” he says.

Cropley says increasing use of the cloud in video surveillance systems can mean changes to the role of systems integrators.

“Software updates and upgrades are currently often part of this role. However, many of these can be provided remotely through the cloud,” he says.

Drako says some of the dealers adapt to this change very well, while others struggle with it.

“The big change is that you have an ongoing relationship with your customer,” he says. “In the historic situation, the customer would call up and say, ‘Hey, we need we need 200 cameras installed and we want AI, and we want this, we want that.’ And you want to bid on this, and the customer will get three bids, pick one of them, a vendor will come out and install all this stuff, wire it up, turn it on, train the customer a little bit and head off. The customer might call them if they want to add a camera, delete a camera.

“But in the case of a dealer or installer providing a cloud system, they’ve got an ongoing relationship with that customer. Every month that customer is paying them to operate the system, because it’s all in the cloud. And the dealer or the reseller can offer additional services. We’ll make sure the cameras are running. We’ll monitor that everything is operational. We’ll automatically come out and fix stuff when a camera breaks. Now the dealer can offer an ongoing service and have an ongoing relationship and take better care of that customer. That’s better for their business, because it means they have a stronger relationship with the customer and a continuous revenue stream.”

Drako says this is also better for the customer because they do not have to deal with the headaches of keeping the system running because the dealer is managing that.

“Eagle Eye takes care of the operation and the upgrades of the software and maintenance, and the dealer can take care of kind of the on-site maintenance — cleaning the lenses, fixing wires that go bad, hanging new cameras for the customer, and maybe making configuration changes for the customer, configuring the analytics or the AI to do what the customer wants,” he says.

When the AI gets involved, the configuration and setup, and the integration into the customer’s ecosystem becomes a lot higher, Drako says.

“If you’re integrating cameras at a construction site, and they detect that someone’s not wearing their safety hat or their safety vest, then the system is notifying somebody. And so, the customer is actually interacting with the system on a daily basis. They’re going to have things that need to be enhanced or adjusted, much more so than in a traditional video surveillance system that’s not using AI. And so, the dealer ends up with a much stronger bond with a customer, more trust and a better long-term business,” he says. “It makes the upgrades, bug fixes and security a lot better, too. The cloud takes away all of those concerns from the customer and the dealer. The dealer doesn’t have to be scared of cybersecurity anymore.”

Maintenance and upgrades

Maintenance and upgrades of cameras have always been a plague of the security industry, but there has been increased interest from clients looking for enhanced cybersecurity.

“In general, once you put a reader on the wall, or a camera in the corner, it stays there until it fails. It’s not an investment centre. For most businesses, they’re not making money on their security system. So, we’ve had to get creative on how we describe the power of surveillance, security, video, access control intrusion, from a return-on-investment standpoint,” Tomasko says. “But the threats today are not the same as the threats we had 15 years ago. It’s not so much that someone’s trying to break through your front door with a hammer or crowbar. They’re trying to break through your digital barrier.”

The number of hacking incidents has spiked since the onset of the COVID-19 coronavirus pandemic, so companies are putting more resources into trying to protect themselves from digital threats.

“That has forced companies to take a bigger interest in keeping their systems up to date, to keep on upgrading, to encouraging installers to say, ‘We need to be at the latest, or near latest firmware or hardware.’ Because cameras from 15 years ago, they may have used nonsecure methods,” Tomasko says. “And if your hardware wasn’t up to spec … that’s a lever to now upgrade. So, we’re seeing a lot more budget and understanding of upgrades from that cyber or IT space, because they have to protect not only their physical walls, but their digital walls as well,” Tomasko says.

DeSousa adds that the long-term implication for maintenance and upgrades to AI-enabled edge-based technology should be minimal.

“As long as you are following current IT and cybersecurity policies, these technologies should already be receiving regular firmware updates. These same updates that carry cyber patches and critical fixes are also the vehicle utilized to deliver updates to AI models,” he says. “These updates can not only offer continue improvements to model accuracy and performance but will sometimes include new AI classifications and capabilities.”

AI drivers

From the user’s standpoint, what are some of the biggest trends driving demands for the different features of AI? They want to do more with less.

“They want to get smarter with how they spend their time. The result of the success of security is we’re not seeing fewer and fewer cameras being installed in businesses, we’re seeing more and more.… Once you install the first camera, that’s the hard one. Once you get one in, you want four, you want 12, you want 100. All you’re thinking about is those areas you’re not seeing,” Tomasko says. “What does that mean? It means it can get overwhelming very quickly for operators or people who have to look at the system.

“If a casino, for example, or a utility, have 5,500 cameras throughout their properties, how do you watch all that video? What AI has given us, not only from a video analytics perspective but also from an entire machine learning and digital perspective behind the video, is more directed events and incidents where if one thing happens, we’re sending screwdrivers. If something else happens, we’re sending security guards and law enforcement.”