Early video analytics, although impressive at the time, were neither accurate nor affordable enough to be accessible to most users. In certain instances, analytics like movement tracking and basic object recognition helped organizations that could afford to implement them, but they were still hampered by limiting factors such as camera quality and processing power. Today, that’s changed significantly, with modern chipsets enabling internet protocol (IP) cameras to effectively and accurately process video natively on the device itself, providing real-time alerts to security teams while sharing vast amounts of metadata with cloud servers for further analysis.
Improvements in machine learning (ML) and deep learning (DL) have been key drivers of this change. The deep learning processing units (DLPUs) that enable analytics to run at the network edge are no longer restricted to high-end devices for wealthy corporations—they’re fast becoming the industry standard. This means that a growing number of customers now have access to advanced AI capabilities, and developers are racing to find exciting new applications to offer them. It’s an exciting time for both users and developers—and it’s largely thanks to the increasing availability of AI.
Deep Learning Is Changing The Game
It’s hard to overstate how important the advent of the DLPU has been. Organizations have long been able to run analytics in the cloud, but this can present problems of its own: Uploading entire video feeds to the cloud can cause significant bandwidth strain, and the cost of storing such a high volume of video can quickly skyrocket. Improvements in video compression have helped alleviate the problem to some extent, but cost is often still a problem—especially for smaller businesses or businesses with a significant number of cameras in use.
By allowing video to be processed on the device itself rather than uploading video to the cloud, DLPUs can dramatically reduce the amount of bandwidth and storage space needed to operate video analytics. Today’s cameras can identify and classify relevant data, sending that relevant metadata (e.g., red car, California license plate, entered parking lot from the west, 2:13 a.m.) to the cloud rather than the entire video.
That data can then be further analyzed: Do vehicles often loiter in the parking lot after hours? Does the same vehicle frequently return? Is the driver trespassing? Behaving suspiciously? Should the police be notified before an incident occurs? This hybrid approach to balancing the edge and the cloud is making those insights available to the growing number of businesses using cameras equipped with DLPUs.
New Applications Give Cameras New Purpose
Not long ago, security applications were the primary use for video analytics. Before analytics, video surveillance systems were typically used for forensically reviewing incidents after the fact to search for evidence or identify the culprit and their methods.
Analytics has helped security sift through hours of video much more efficiently, and in some cases, make security more proactive by issuing real-time alerts when suspicious activity is detected. Sometimes, this can stop a crime in progress. Other times, it may allow security personnel to defuse a potentially dangerous situation before it has time to escalate into an emergency. From retailers looking to reduce theft to a remote oil refinery seeking to identify trespassers, these analytics have obvious security applications.
But as deep learning becomes standard in many cameras, users and security departments have become increasingly interested in how they might use those capabilities in ways that go beyond security. If a camera can track the movement of people and vehicles, after all, couldn’t that information provide valuable business intelligence? Today’s businesses want more data, and video analytics can help them better understand their customers.
For example, a retailer might use video analytics to track when customers are most likely to visit their store, how they travel there, where they go first upon entering and which displays draw their attention. They can then use that data to inform staffing decisions, the store layout and future promotions. Best of all, customers don’t even need to purchase new hardware—they can gather these insights using the same devices they’re already using for security.
The growing demand for video analytics has spurred a corresponding influx of new developers looking to take advantage. It isn’t just a matter of security analytics developers pivoting to more business- and operations-related applications—entirely new groups of developers are now building video analytics solutions.
With AI no longer limited to those with significant resources, developers are identifying opportunities in the market and creating specific analytics designed to meet them. This might include analytics designed to assist with quality assurance, helping manufacturers identify defects before they leave the factory floor, or others designed to identify bottlenecks and other inefficiencies at retail stores, event venues and other locations. The growing availability of deep learning technology has sparked a veritable gold rush of development, and users are reaping the benefits.
Using Technology To The Fullest
The best part about the increasing ubiquity of AI is that many organizations may not even realize the advanced new capabilities they have at their fingertips. Cameras currently being used exclusively for security purposes may be able to do far, far more, and organizations should work with integrators—or directly with manufacturers—to ascertain whether the cameras they currently have in use contain DLPUs and how they can put them to better use.
Many organizations may learn that their existing surveillance setup can not only keep their physical locations secure but also provide them with the actionable insights they need to move their businesses forward. As the use of AI continues to expand, new applications will continue to arise—and today’s businesses should be prepared to take advantage.