As end users seek to meet their security needs and requirements, they increasingly turn to AI, prompting start-ups and established companies alike to develop cutting-edge solutions.
As end users seek to meet their security needs and requirements, they increasingly turn to AI
, prompting start-ups and established companies alike to develop cutting-edge solutions.
In security, end users are becoming more proactive and want to detect suspicious or blacklisted individuals early on to prevent incidents. To pinpoint them from thousands of video feeds, AI-based video analytics can be useful in the interest of time-saving and efficiency. In particular, face detection and recognition, a lot of them AI-based, have become more common, seeing increased deployment in airports, offices, factories and other end user organizations.
For facial recognition
, accuracy is the name of the game. Especially in challenging conditions, for example areas with poor lighting or when faces are in motion, solutions that can still accurately detect and recognize faces can win out amid the competition. One company that stands out in this regard is SAFR, which leverages deep learning to turn video streams into actionable intelligence with what the company claims to be 99.87 percent accuracy.
“SAFR is the most accurate facial recognition solution for live video available in the market today. We make this claim based on the findings from the National Institute of Standards Technology (NIST) … whose latest test evaluated 185 algorithms from across the world and once again found SAFR to be in the cohort of most-accurate algorithms for wild images,” said Eric Hess, Senior Director of Product Management at SAFR.
SAFR can benefit any user who needs facial recognition to identify people on a watch list and get accurate, instant alerts. “Some verticals with large watch lists where we’re seeing traction are airports and other transit centers, law enforcement, and casinos and gaming,” Hess said.
Common camera platform
More and more, AI-based analytics or apps are installed on the edge instead of on the server to shorten response time and optimize bandwidth. However, the camera operating system market is quite fragmented; the developer has to develop different versions of apps to fit different camera brands and OS’s. Offerings to systems integrators and end users are fragmented as well, making it hard for them to look for what they need.
That said, a common camera
platform can be an effective solution, allowing developers to develop apps only once for different camera brands. In this sense, Security and Safety Things is an innovator, offering an open camera operating system as well as an app store through which different camera apps supported by the camera OS can be obtained.
“End users can pick and choose from a broad selection of video analytics applications in our application store, so that a camera can for example simultaneously detect a person of interest, the formation of a crowd or the presence of a weapon in a person’s hand,” said Fabio Marti, Director of Sales and Marketing at Security and Safety Things. “By creating value at the camera level and enabling these AI applications, we are hoping to focus the industry’s attention back on increasing intelligence and value at the camera level.”
Marti further notes that currently, the operating system features more than 40 ready-to-use apps. “The cameras will be available beginning in 2020 from a variety of manufacturers and their distribution channels – systems integrators, distributors or other existing sales partners,” he said.
Increasingly, home security devices such as IP cams and facial recognition-enabled smart locks are also seeing demands for AI to improve accuracy and precision. However the market is lacking AI-enabled, high-performance and low-power systems-on-chips
(SoCs) for these devices, a lot of which only run on battery power. In this regard, San Diego-based Kneron fills the gap.
"What we deliver is edge SoC that is capable of running CNN-trained AI algorithms at low power. Most edge devices have limited computing resources to run neural networks. Our SoC on the other hand can run complex neural networks at less than 500 milliwatt,” said David Yang, Special Assistance to CEO at Kneron.
The company, whose markets include Europe, North America, China, Southeast Asia, Japan, Korea and Taiwan, primarily targets smart home devices. One example is smart locks. “We see increasing adoption in smart locks as they evolve from keypad-based to fingerprint-based to now facial recognition-based. With our solution, even these battery-powered locks can run complex CNN-trained facial recognition. This is our advantage,” Yang said.
IP cameras are another application. “With our edge SoC, cameras can filter out unnecessary information and upload only video streams that are meaningful, for example those with human presence or a specific behavior in it,” he said.
According to Yang, especially in Europe and North America, a lot of people balk at the thought of AI, citing privacy concerns. “Our solutions are for edge devices so there’s little privacy issues. Yet some end users are still concerned. We’ll step up communications with our customers and persuade them with our proven record,” he said.
Johnson Controls, which has launched its own Tyco AI deep learning solution, also mentions the importance of education and privacy-by-design practices. “Education really is key to ease concerns about privacy and data protection. As an industry, it’s important that security professionals work together to reassure people that information collected remains secured and will not be used inappropriately,” said Ric Wilton, Director of Product Management at Johnson Controls. “Because of technology we now have access to many great tools that enable systems to more accurately and quickly identify objects and people of interest. To address privacy concerns we have developed facial masking technology that enables users to leverage privacy constraints when needed. All of our solutions also go through rigorous cyber security testing.”