For at least the last couple of years, artificial intelligence (AI) has been one of the biggest trends in the security industry. Offering the ability to “learn” over time, AI provides the technology to anticipate and identify potential issues as they are unfolding. Given the value of video in security, it was a natural fit for AI applications, making a marriage between the two inevitable.

In fact, the growth and development of AI for video can be traced to the evolution of video analytics over the last few years — to the point where analytics and AI are so intertwined it’s almost difficult to tell where one ends and the other begins. Regardless, the combination has tremendous impact on the industry.

“Deep learning has become a very large, transformational force within the security industry,” says Eric Hess, senior director of product management, SAFR, Seattle. “It has allowed computer vision to advance to a point where it can approach many of the tasks that a human would do, just by simply watching a video feed and being able to recognize with excellent performance when a particular incident or action visible within a video feed should be reviewed by a human. And that’s probably one of the very important elements to understand here: that as good as deep learning and AI is — and in many cases, there are instances where something like face recognition can actually be done more accurately and with less bias by the computer than by a person — it’s still as such that humans need to review the results and decide the next course of action. But it definitely has been a transformational force for the industry.”

One example of how video analytics has fueled the development of AI can be seen in the evolution of facial recognition, says Greg Cortina, director of sales, Dahua Technology USA, Irvine, Calif.

“Facial detection captures facial images as snapshots and cannot identify individuals; it’s just a method of collecting images for later review by the user,” he says. “But once those images have been categorized as faces, facial recognition can create positive IDs of individuals through the use of deep learning algorithms.”

Advancements in the reliability of video analytics have made machine learning possible, says Paul Garms, director of regional marketing, Bosch Security and Safety Systems, Fairport, N.Y. Machine learning is part of narrow AI and is a way of training an algorithm by feeding it data as a method of teaching it to adjust itself to improve its performance.


Are Analytics Accessible for SMBs?

For enterprise-level companies and organizations, video analytics are not only accessible but are often deployed as part of a standard video surveillance and security deployment. But for the SMB market, where budgets are small and margins are often smaller, that type of technology hasn’t always been attainable.

“The main stumbling blocks for AI adoption in security applications have been processing power and cost. Initial solutions involved complex server-based setups and specialist services, putting them out of reach of most organizations,” says Jason Burrows of IDIS. “At the same time, there were limited real-world application references or education programs that clearly demonstrated return on investment.”

Today, Burrows says, price points have fallen to a place where AI functionality has become accessible through existing VMS solutions and via cloud-based applications, with vendors presenting more clear and concise applications that address everyday challenges.

“However, it’s still critical to build business cases that demonstrate ROI,” he says. “Manufacturers and systems integrators need to help customers understand their biggest pain points and challenges and show how they can operate more efficiently, while continuing to reduce risks, increase safety, and meet compliance and duty-of-care requirements. If we do that, we’ll see greater adoption as users find many practical applications, including some really useful ones that we haven’t yet envisaged.”


“It enables integrators to tailor video analytics to detect the things that matter most to their customers,” Garms says. “The camera retains information on user-defined objects and situations and refers to these learnings when processing scenes. This gives integrators more flexibility in the applications of video analytics for their customers.”

To properly understand the role of AI in video analytics for security applications, you have to look at the end goal: to drive video operators’ focus from unstructured, seemingly endless video streams toward informative metadata that motivates action and generates intelligence.

“This means taking advances in computer vision, machine learning and pattern recognition, and adapting them to a security context … that facilitates rapid responses and augments human resources,” says Tom Hofer, product manager, Senstar, Ottawa, Ontario.


From Reactive to Proactive

Looking at the evolution of analytics means starting out by looking at the evolution of cameras themselves, says Alex Walthers, business development manager – application development partner program, Axis Communications, Chelmsford, Mass. Traditionally, cameras were deployed as a deterrent, with mixed results. From there, cameras became forensic tools for investigating incidents after the fact. And as video analytics have evolved, cameras and systems have moved from forensic tools toward more proactive solutions.The combination of analytics and AI promises to push this evolution even farther, Walthers says.

“With the development of AI, we’ve seen that you can more thoroughly and accurately identify human behaviors and human characteristics to determine if something is an animal or if it’s a criminal in an area they’re not supposed to be in,” Walthers continues. “Using AI, video cameras can be trained much more quickly on thousands and thousands of images to continue to build upon and improve itself.”

Tasks like people counting, line-crossing and perimeter detection are among the narrow-focus, repetitive tasks analytics excel at performing — far better and far more efficiently than a human could even dream of.

“A lot of the repetitive tasks that would have traditionally been done by a person are now freed up so those professionals can take the data that’s generated and do something with it — interpret it, respond to it, interact with it in some way,” Walthers says. “So we’ve really seen that shift from deterrent camera to forensic search features to proactive response to improve performance, to protect an area or to increase operational efficiencies for business.”

According to Florian Matusek, product group director – video analytics, Genetec, Montreal, AI is the latest plateau video analytics have reached in their ongoing evolution.

“Machine learning is another plateau where you might say 10 or 15 years ago, it was tracking objects. That was maybe a step ahead. Nowadays, it’s machine learning and being able to do the same applications analytics did 15 years ago but much, much more accurate and really usable. This has really triggered the revolution in our space,” he says.


The ‘Game Changer’

If analytics moved the needle in improving security, AI will take that improvement a lot farther, says Sean Foley, senior vice president, Interface Security Systems, Earth City, Mo.

“AI is really the game changer here,” he says. “When it comes to asset protection technologies, it’s going to really cause a parabolic shift in the types of technologies our clients are using, the type of solutions that we provide to them, and the level of effectiveness that they have in combating all of the things that loss prevention and asset protection teams are dealing with on a daily basis, namely, the security of their customers, their associates and the reduction in shrink.”

Two of the biggest factors that can create challenges for video analytics are video quality and what appears in the camera’s field of vision. Too much movement or low-
quality video can make it difficult for the analytics to determine exactly what is happening, causing false alerts. However, AI has the power to change that, making analytics much more reliable, says Aaron Saks, product and technical manager, Hanwha Techwin America, Teaneck, N.J.


Terminology Confusion

Regardless of its ability to improve video analytics, one of the problems with AI is the term itself, says Genetec’s Florian Matusek. While the name indicates “intelligence,” that isn’t necessarily the truth.

“AI really means automation. So usually when we want to talk to you today about AI, rather than AI, we say IA, or intelligent automation. What it really does is it automates stuff; but it does it very well by using machine learning and deep learning. But calling it AI suggests that it’s super intelligent, even close to a human, which it is not,” he says.


“The issue is that for most or many of these types of analytics, the proof of concept was out there, how it should work, but it doesn’t always work perfectly. And the installation environment, you have trees, windows, cars driving by — you have too much noise, messing up the information that you want,” he says. “AI kind of grew out of that, as we have more of these hyper-intelligent sensors or processors, edge devices or cloud or servers that can better refine and filter out the garbage, filter out the noise, so you can see what the actual data is through the forest of trees so to speak.”

Foley says that on the AI front, and for asset protection teams in particular, actionable intelligence is the name of the game. They need good data, but they also need data they can act on, and preferably that action can come sooner, rather than later.

“This is really facilitated by solutions that vacuum up and process as much data from the store environment as possible,” Foley says. “This can be everything from solutions that monitor transactions at the point of sale; solutions that monitor video; alarm event data; when a store opens or closes; when motion is happening in certain parts of a store environment; and all the more of that data that you can put into a single repository and process that.”

Video analytics and/or AI also have the power to impact current video deployments used for simple monitoring operations, making them capable of detecting problems as they are unfolding and providing users with potential solutions quickly.

“For years cameras have monitored major highways and intersections and used simple analytic triggers,” says Jason Burrows, regional sales director, Western U.S., IDIS, Austin, Texas. “Now, we’re seeing the benefits of intelligent traffic flow systems, for example automating variable speed restrictions and traffic light patterns to prevent roads from becoming gridlocked or alerting drivers to obstructions ahead.”

Finally, we can expect AI applications to continue to evolve as vendors look to decentralize computation away from servers and the cloud to the edge of the network. This will mean lower bandwidth requirements, faster alerts, and most importantly even lower price points.

This is where AI combined with video analytics comes in, delivering the kind of intelligence that allows organizations to either act quickly to mitigate an incident or, ideally, stop it before it can even occur.

The future is incredibly bright, Foley says. “AI is helping us with the process. It’s simply incredible what you can do with artificial intelligence and where it’s going. And frankly, in our space as in so many others, we’re kind of just scratching the surface,” he says.



Objective of the SDM 100

For more information on this topic visit SDM’s website, where you will find the following articles:


“Video Analytics: Mature Solution or Still-Developing Technology?”


“State of the Market: Video Surveillance”