As with nearly every other sector of the economy, the physical security industry has been abuzz of late when it comes to AI and how it will impact not only video surveillance but also the market more broadly. However, unlike industries that are just becoming acquainted with AI through things like ChatGPT – the technology is already being successfully used in a variety of applications such as surveillance to make security more efficient.
Though estimates vary, multiple market research firms predict that demand for AI-powered video analytics will grow exponentially over the next decade. According to Fortune Business Insights, for example, the video analytics market is projected to grow from just over $6 billion in 2022 to over $28 billion by 2029. Another report published by Verified Market Research forecasts the market to grow from $5.65 billion in 2021 to approximately $36 billion by 2030.
The fact is that the security market finds itself at a technological inflection point when it comes to AI and video analytics in what many see as a “third wave” of video surveillance innovation with analog cameras and digital video recorders (DVRs) constituting the first wave and network cameras and video management systems (VMS) comprising the second. Similar to the move from the first to the second wave, there will certainly be leaders and laggards in the current transition and where organizations fall on this scale will undoubtedly go a long way in determining how they are positioned for the future.
Just as there are a fair share of skeptics surrounding AI today, there were also a number of vendors who felt that IP would not overtake analog to the degree it has today. It remains to be seen exactly when video analytics will become an industry norm like IP or HD resolution, but rest assured it is only a matter of time.
The unprecedented demand for AI in security is being driven by operational efficiency – the need to eliminate false alarms and thereby reduce the time that operators, guards and other employees spend responding to them on a daily basis.
Of course, the need for false alarm reduction has been near the top of every security end user’s technology wish list for decades. Monitoring and cloud-based systems have only accelerated this demand to drive false alarm reduction. However, it was only recently that new innovations in object recognition and classification, driven by neural network training, have made this capability a reality.
It’s important, when talking about these systems, that we agree on definitions. Though many people consider true AI to be the ability of machines to analyze data and make decisions without any additional input from humans, when it comes to security and video surveillance, AI means being able to trust analytics as if it were a live guard providing the notification.
Typically, the analytics used in security applications leverage algorithms that have been trained using either traditional machine learning or deep learning methodologies. With standard machine learning, computers are given small amounts of data so they can identify simple patterns or trends within video. Conversely, deep learning (or neural network training) involves using large amounts of data to train computers to process data much in the same way a human brain would.
Today, leveraging these evolved systems of analytics, end users across the board can detect people, animals, vehicles and other objects with extremely high degrees of accuracy. At ISS, for example, each of our analytic modules deliver accuracy rates exceeding 90% in most instances, with most running at 98% or better.
In addition to improvements in the technology itself, there are also larger economic and societal changes that have taken place over the past two decades that are spurring not just a desire to leverage video intelligence but have necessitated their deployment. The digital transformation that is taking place today, for example, with organizations adopting modern technologies to improve their business and bottom line means that video surveillance systems are no longer seen as solely as a tool to be used for post-incident investigations.
While most end users want to simply be able to determine whether an alarm was triggered by an actual person or if it was just foliage blowing in the wind, the advent of surveillance analytics has provided organizations with the ability to collect and aggregate a variety of different datapoints to not only improve security, but to also increase efficiencies across a business.
In the hospitality sector, for instance, hoteliers can now use algorithms trained on things like crowd detection to determine how long people have been congregating in certain areas and subsequently allocate personnel to address any issues that might be holding up guests checking in or out. Marketing professionals in retail can also use dwell time and heat mapping analytics to discern what displays are the most effective and which ones fall flat.
Another area where video analytics will play an invaluable role moving forward is in workplace safety and regulatory compliance. While many people around the world became intimately familiar with donning personal protective equipment (PPE) during the COVID-19 pandemic, for others, continuously wearing PPE is just another routine day at the office. In fact, within certain industries, the wearing of things like hard hats and safety vests is not only a good business practice but is mandated by various regulatory authorities.
According to the US Occupational Safety and Health Administration (OSHA), about 1,000 workers die every year due to head injuries sustained on the job and of these deaths, more than 80% are suffered by those not wearing a helmet. OSHA fines issued to companies for employees that fail to wear hard hats can also result in significant costs for businesses. From October 2021 through September 2022, the US Department of Labor reported that just over 840 citations, totaling nearly $3,000 apiece, were issued to companies that ran afoul of OSHA’s rules pertaining to wearing head protection on the job.
Trying to keep track of these incidents with the traditional search and archiving functionality of a traditional VMS would be practically impossible; however, leveraging analytics specifically trained to monitor and alert to the presence or absence of PPE in conjunction with a modern video intelligence platform would enable organizations to track, in detail, every violation and address it promptly with both employees and regulators.
In addition to hard hats and safety vests, analytics can also be trained to recognize the presence of other PPE, such as face masks in healthcare settings, or even whether workers have performed safety-related tasks like washing hands prior to entering a food preparation space within an agricultural processing facility. Regardless of what a business wants to keep track of, analytics can be created and trained to accurately monitor it.
As anyone who has been around the security industry for any length of time can attest, the high degrees of accuracy that are capable with today’s AI-powered analytics were the stuff of science fiction when the technology made its first go-round in surveillance market more than a decade ago. In the late 2000s, much like today, there was a flood of companies entering the market making claims about the capabilities of their analytics which largely failed to hold up under real scrutiny. This age of overpromising and underdelivering left a sour taste in the mouths of many integrators and end users, who were essentially sold empty boxes.
Granted, there are still companies making bold claims about their analytic capabilities today, but there are ways to separate the wheat from chaff when it comes to selecting an AI partner.
One of the first things you should ask is how long has the company been around? Are you comfortable placing your trust in a company that has only been around for a couple of years without any patents or intellectual property to their name? They may be flush with venture capital and have a sleek marketing presentation, but will their technology stand up in a real-world deployment once it is brought out of the lab? In many cases this is not only a financial consideration, but in terms of neural network training, limited time in business can result in limited datasets to train an analytic on.
Also, you need to determine what kind of accuracy the analytics provider you are evaluating can achieve on a consistent basis. Are the analytics high trust, meaning would you accept an alarm generated by the technology the same as you would a human sitting in their place? What kind of datasets were their analytics trained on? Were the algorithms shown a handful images or were they presented with hundreds or even thousands of hours of footage to be able accurately detect realistic behaviors?
It is up to you as consumers and installers of these solutions to evaluate their effectiveness and avoid the pitfalls of the past. AI in surveillance is here to stay – the only question is when and how you decide to take advantage.
Matt Powell is Managing Director for North America at ISS (Intelligent Security Systems), a pioneer and leader in the development of video intelligence and data awareness solutions. He has over two decades of experience in security and transportation technologies, having formerly served as Principal-Infrastructure Markets at systems integrator Convergint and as a developer of transportation market strategies for Videolarm and Moog prior to that. He can be reached at [email protected].
This article was originally published in the May edition of Security Journal Americas. To read your FREE digital edition, click here.