Artificial intelligence (AI), deep learning and machine learning are now common buzzwords used across numerous disciplines and vertical markets. But what’s the reality as far as its current applicability in physical security? According to the experts, we’re entering into the realm of providing real-world applications and generating pertinent data that assists with higher levels of security and safety.
“AI and deep learning as it applies to the security industry is a platform for analytics. Fundamentally, the technology is about making analytics more accurate, capable and powerful,” said Oliver Philippou, research manager, Security Technology, IHS Markit, London.
He said one of the key opportunities for deep learning is video surveillance and the ability to recognize and identify images.
“Currently, AI and deep learning falls into several primary categories, including vehicle recognition, facial recognition, behavior recognition, object recognition and video search and synopsis. Let’s say you are looking for a person of interest who drove on a particular road in the afternoon. If you know the car is blue and traveled down that road, it can instantly filter the video to get the images you are looking for.”
On school campuses and airports, cameras with analytics and AI can identify a suspicious package and alert authorities in real time. Retail stores can leverage analytics to prevent potential loss or fraud and personalize the consumer marketing experience from data gleaned from store visits. Smart video for municipalities assists in transit or citywide operations, bringing greater situational awareness. With more cameras being installed and video streams watched, AI uses self-learning video analytics that automatically identify potential issues, so operation center personnel can pinpoint actual alerts or events quickly and proactively.
The speed of development of AI has propelled its adoption, particularly in the case of deep-learning algorithms in the video surveillance market, according to the white paper “AI in Physical Security,” produced by the Security Industry Association, Silver Spring, Md., in partnership with IHS Markit.
Today’s deep-learning analytics, the report states, offer benefits in accuracy and power:
“The ability of deep-learning algorithms to view a scene intuitively as a human would means that detection accuracy increases dramatically. Also, as computer power continues to increase, neural networks will leverage this to process more data and improve accuracy, which is an important development for the video analytics industry”
In addition, a combination of more powerful GPUs and the ability for analytics to automatically detect, recognize and classify objects has made video searchable, the report states.
“The AI-driven algorithms utilizing deep learning techniques are already there as far as capabilities,” said A.J. Frazer, vice president, business development for Agent Vi, New York. “The next challenge is how to build the data and compute infrastructure to support these complex algorithms. Deep learning requires a lot of data and levels of GPU computing that our industry is not yet used to.”
Unlike other technologies and software tools, AI depends heavily on specialized processors. To meet the complex demands, manufacturers are creating specialized chips. A growing trend of AI is putting the “smarts and parts” in the cameras, but one of the downsides has been thermal issues or heating that results in choppiness, which is a hardware limitation. This thermal-heavy processing has been a limiting factor in implementation, but new chipsets and processes—such as CVflow computer-vision architecture that provides lower power processing developed by Ambarella, Santa Clara, Calif.—are overcoming these limitations.
Frazer said Agent Vi’s open architecture video-analytics solution, as a cloud-hosted software as a service, can handle the data and computing challenge of AI. The company’s innoVi solution uses deep-learning techniques that are trained on real-world surveillance and not stock footage, which allows for an extremely high level of accuracy. Agent Vi was the security industry’s first video analytics software company to commercially deploy advanced deep learning solutions.
AI-driven video analytics can be trained to detect bicycles and motorcycles, so that alerts are issued when a motorcycle is seen on a bike path, and it can detect and alert sanitation services to overflowing dumpsters for streamlining trash removal.
“In the cloud, data is easy to share and computer resources can be turned up as needed,” Frazer said. In addition to smart cities and other large installations, AI can be purposed beyond security, for traffic monitoring, waste management and other services.”
