AI • IOT • NETWORK EDGE

Despite widespread use of video-based security, until recently its effectiveness has been hit or miss. For example, property owners usually count on receiving alerts when a person crosses a secured perimeter, but an animal could trip those sensors just as easily. As a result, operators often have to sort through false alarms—and manually search hours of recorded video—to find events that mattered.

But AI-based video analytics are changing everything. Using sophisticated deep-learning algorithms, these edge-to-cloud systems watch and analyze video, filter the noise, and inform operators in real time when action is needed.

As a result, security staff can be prepared and ready to respond to valid alerts as soon as they come in. And instead of spending hours “rounding” the premises, personnel can perform more valuable tasks—like interacting with customers.

But improved security is only one benefit of this new technology. By relying on edge processing and limiting the use of personal identifying information, its design also ensures responsible use of surveillance video.

Harnessing the Power of Video

The main advantage of AI-based video analytics over human viewing—and even traditional computer vision—is improved accuracy. A license plate recognition use case is illustrative: Numbers and characters might be vertical or horizontal, and colors, size, symbols, and more differ by state or country.

Traditional computer vision systems can’t surpass 40-50% accuracy in this situation, according to Prush Palanichamy, Vice President of Sales at Uncanny Vision—an AI video analytics solutions provider. And that kind of performance renders the system effectively useless.

Uncanny’s AI-based vision systems, on the other hand, are guaranteed to be 95% accurate. That’s when they really start to become useful, according to Navanee Sundaramoorthy, Co-founder of Uncanny Vision. But site-specific training can raise that figure even further—to 97 or 98%. “And then the system becomes so easy to use, and so effective, customers wouldn’t think of going back to earlier methods,” he says.

But this level of precision has nothing to do with collecting personal identifying information. On the contrary, the system doesn’t care about faces or license plate numbers. Images are rarely even stored, let alone sent to the cloud—which would be cost-prohibitive for both cities and businesses.

Instead, Uncanny converts video into metadata at the edge. Processing video locally means only a few kilobytes of data need to be sent to the cloud for analysis, which serves as the basis for automatically generated alarms. This way customers save money as well as the trouble of securing confidential information.

With @UncannyVisionAI’s help, SPs can turn a city’s existing #CCTV cameras into cost-effective vision #sensors, improving the operation of entertainment venues, office buildings, and transport hubs. via @insightdottech

Applications for AI-based Vision Are Endless

A customizable framework built with the Intel^® OpenVINO^™ Toolkit on small embedded processors at the edge makes that kind of processing possible. In fact, according to Sundaramoorthy, OpenVINO optimization improves performance by four to eight times. This means the system can process video at 30 to 40 frames per second instead of 10, and it can be used for a whole host of applications—like highway traffic monitoring—that wouldn’t be possible without such acceleration.

Those applications—big and small—are making public and private life better all the time. For example, some service providers (SP) are already using Uncanny’s solutions to provide additional value to cities on top of the high bandwidth connectivity they’re already delivering.

With Uncanny’s help, SPs can turn a city’s existing CCTV cameras into cost-effective vision sensors, whose insights can improve the operation—and beauty—of entertainment venues, office buildings, and transport hubs.

But even more critical, roadway safety can be increased—by using data on what’s happening here and now to plan traffic activities, instead of a 10-year-old traffic survey. Sundaramoorthy points out that real-time insights are especially important today, given how the pandemic has changed how we live, work, and travel. “Traffic patterns have changed so dramatically in so many cities,” he says, “historical data is practically useless.”

The only way to keep up is with a continuous, up-to-date view of all the relevant metrics: the number of vehicles, pedestrians, and cyclists on the road—as well as their location, direction, and speed. This was the motivation for Uncanny’s traffic analysis use case, which comes with a dashboard that tells cities everything they need to know.

And AI-based video analytics are keeping highways safe, too. For example, the U.S. Department of Transportation is using Uncanny’s solution to let truckers know where they can find available overnight parking spaces. So instead of stopping on the side of the road and jeopardizing their own and other drivers’ safety, they can pull over whenever they need to—without wasting time and gas looking for a spot.

New Opportunities for Systems Integrators

Uncanny makes it easy for cities and organizations to transform their security cameras into smart vision sensors. Typically, local systems integrators (SIs) install the system, and a successful deployment doesn’t require any special knowledge of AI.

“Anyone who knows how to install a CCTV camera can install our system,” says Sundaramoorthy. Uncanny works with various companies—including Lenovo, Asus, and Ingram Micro—which in turn load the Uncanny software onto their Intel^® processor-based systems. All that remains is to connect the CCTV camera to one of these smart boxes.

That means this technology could be a boon for SIs, too. In addition to their domain expertise, now they can bring their customers a subscription-based, camera-plus-analytics service. This higher-value offering could serve as a new, steady revenue stream, raise SIs value to their customers, and help them transition to a more competitive business model.

The benefits of AI-based video analytics systems extend in every direction. Enhanced performance for the people who use them. Resource savings—time, money, and human capacity—for organizations. And better quality of life for the rest of us. All the hallmarks of a revolutionary technology we can’t afford to ignore.