Video surveillance evolves from 'dumb' cameras to 'intelligent' sensors

The video surveillance industry it seems has been enamored with the potential of moving beyond its traditional purview of security going back nearly a decade. While the industry’s initial foray into analytics resulted in technology that largely overpromised and under-delivered, it did lay the groundwork for today’s artificial intelligence (AI) and machine learning technology which has subsequently opened the doorway to a new wave of market entrants and solutions. As with the first analytics wave, time will tell which of these companies and technologies have industry staying power, but it’s clear that providing a deeper level of actionable video intelligence is going to be a requirement for next-generation surveillance systems.    

In many ways, the path of today’s video surveillance products parallels that of the residential alarm technology, which has undergone a transformation in recent years from being simple intrusion detection systems to whole home automation solutions. Similarly, video surveillance is no longer seen as just a method to keep people and property safe, but also as a tool to help improve business operations.

Unlike the migration from analog to IP, however; most vendors today are not content to rest on their laurels and wait for this technology evolution to pass them by. Nearly everywhere you turned at ISC West in Las Vegas last week, video surveillance manufacturers were touting the additional intelligence their solutions could deliver and, most importantly, how it is going to put more money in the pockets of integrators and their end-user customers.

Here is brief snapshot of some of the current industry trends that were on display during ISC West 2018 and how they are impacting the product development roadmaps of various companies.            

Artificial Intelligence (AI) Permeates the Industry

Perhaps not since the first generation of video analytics hit the market has a technology so thoroughly captured the imaginations of the industry more than AI and machine learning. The thought of computers being able to analyze video by themselves and alert users of only those events that are pertinent to their operations is attractive for a multitude of reasons. But while adoption and case studies of real world deployments are few and far between in North America that could soon change as more companies introduce products that offer the technology as an embedded feature.

In addition to hosting an educational session at ISC West 2018 focused on how “AI Creates Value,” Dahua also introduced its “deep sense” series of products, which includes cameras, NVRs, servers, etc., all of which offer advanced video intelligence capabilities, such as human characteristics analysis, face recognition, ANPR, metadata collection and analysis, people counting, image search, facial flow, traffic incident detection and traffic data statistics.

According to Tim Shen, marketing director for Dahua Technology USA, the goal of AI is to make data better and more actionable for end-users, which is what the company aims to do with its latest suite of products. “People get (what AI is), but they still need to know how this is going to help them,” he says.

Dahua was just one of a number of manufacturers on the show floor last week who were offering more intelligent video solutions. Hikvision, for example, has even incorporated deep learning facial recognition into its new Swing Barrier Turnstile, replacing card readers with a Facial Recognition Terminal for easier access control.

Deep Learning Advances

According to Leon Snyman, Applications Engineer for Video Analytics/Perimeter Security at IDIS, his company has achieved about a 95 percent accuracy level when it comes to object classification with its deep learning algorithms by training the technology on more than six million various images of humans, vehicles, etc., to help it distinguish between objects that pose a threat and those that do not.  

“In just the last six months we’ve seen that improve. Previously it needed to see the whole person, meaning head, shoulders, body, and limbs, but now it can recognize a human just based on the head and shoulders or just a leg, so it has improved a lot,” Snyman says. “It’s going to be a module in our VMS and classification is the most important part. With previous video analytics you were looking at a lot of pixels and you had to interpret what is that blob of pixels. Some solutions went so far as to say, ‘well if it’s vertical in nature, then that’s a human or, if it is rectangular, that’s a car’ but you don’t really know. In this case, you show it a picture and it makes a judgement about whether it is a car or a human.”

Convolutional Neural Networks vs. Spiking Neural Networks

Unlike traditional Convolutional Neural Networks (CNNs) that most Deep Learning technologies leverage and that involve training a computer to recognize certain images, BrainChip is bringing a completely different way of doing things to the industry. BrainChip’s technology uses Spiking Neural Networks (SNN) which enables a machine to learn on the fly without have to ingest large volumes of data for days, weeks or even months in advance.

“When I feed (our solution) data, it’s a single image and then the system responds by training itself, in our case through reinforcement and inhibition of synaptic connections,” Bob Beachler, Senior VP of Marketing and Business Development at BrainChip, explains. “With CNNs and Deep Learning, what they do is take the data, look at the result and they do something called back propagation where they go back and change the weights in the math, do it again through successive approximation where they have the engine setup based upon the data. Once you have it done, the goodness of network becomes a function of how good is the architecture of the neural network, the training methodology and the data set.”

Integrating the IoT with Video

Aside from the improving the intelligence of video hardware and software, another approach some vendors are taking to provide end-users with better data is to simply integrate more sensors into their video management systems. March Networks, which has traditionally been strong player in the banking and transit verticals, is one such company and has found success with its Searchlight solution.

“We started a few years ago going down this path of using video analytics for people counting, dwell time, etc., but customers came back and said it’s not accurate,” Dan Cremins, Global Leader for Product Management at March Networks explains. “So, we started integrating other types of devices and we said, ‘you know what, let’s look at other technologies – they don’t have to be video based – that supplement video to provide meaningful information.’”

Cremins says they started off using 3D stereographic sensors and applied those to video and eventually developed Searchlight, which serves as a video data hub for end-users.

“So many of these catchphrases – IoT, Big Data, Machine Learning, AI –it all comes together. The way I explain it, is that video is just another source of data. When you put video and IoT (sensors) together, that creates Big Data, so then Big Data is honestly useless until you cull through it and get that information and that’s where analytics come into play, taking Big Data and converting it into smart data.” Cremins says. “The last phase involves taking this information, that smart data, and presenting it to all of these other entities or departments (within an organization).”

Traditional Vendors Move Beyond Cameras

Just as the industry has become more focused on delivering smarter surveillance products, companies that were once known predominantly for their high-quality cameras have started to put their R&D dollars and efforts towards a wider array of solutions in recent times. Axis Communications, for example, has rolled out a variety of IP-enabled products at ISC West over the past several years, including speakers, a perimeter intrusion radar unit and, just last week, an audio server. Additionally, after focusing mainly on cameras for much of its existence, Arecont Vision last week introduced a complete end-to-end video solution that includes video management software, video recorders and web services.

Another company that has looked to broaden its portfolio is Hanwha Techwin America. Tom Cook, the company’s VP of Sales for North America, says that after the company was acquired by Hanwha about two and a half years ago, they made a concerted effort to expand their product line and “aggressively grow” their market share, which has paid dividends for the company.

“We’ve recognized that continued growth can’t happen with just being a camera manufacturer,” Cook says.

One of the products that Cook says is indicative of how they’re growing their product portfolio is the recently launched Wisenet WAVE video management system. The company conducted a soft launch of the VMS at ASIS 2017 and formally released it in January.

“It’s a true open platform VMS,” he adds. “One thing you don’t want to do is be a ‘me too’ company. We analyzed the market a lot with what we wanted to do and the company that we’re partnering with, Network Optix, had a completely different approach. It’s a media server and not the typical VMS with the rigidity that is built into a lot of them because of how they were built years ago on older technology.”      

About the Author:

Joel Griffin is the Editor-in-Chief of SecurityInfoWatch.com and a veteran security journalist. You can reach him at [email protected].