Page 12 - Security Today, March/April 2024

ChaoticDesignStudio/stock.adobe.com
Some New Tricks AI-based cameras have some surprises up their sleeve
By Adam Lowerstein
In the constantly evolving world of AI, you can blink and miss an innovation. This rapid pace of evolution means organiza- tions are under increased pressure to invest in solutions that do not become obsolete a short time after they are installed.
For video security in particular, AI evolutions in machine and deep learning are consistently bringing new products and ser- vices to help us do a better job of protecting people and assets. Like our smartphones, features and functions can evolve rapidly when we consider AI apps running on edge devices such as a camera. An app-based ecosystem allows for flexible customization, updates and deployment of features tailored exactly to the job at hand.
For this reason, it has never been more important for compa- nies to seek out vendors and manufacturers with open platforms that collaborate freely with multiple third-party manufacturers.
AI has fundamentally changed video security. We have gone from cameras that can sense basic motion, which were prone to false positives from passing shadows or wind-blown trees, to highly accurate human and vehicle detection with descriptive attributes and real-time alerts in just a couple of years. Further evolution has led to cameras that can detect anomalies in a scene (scene change detection) such as a door being left open, or a vehi- cle left in a no-parking spot beyond a preset time limit. The same analytics can even notify when stock is running low on shelves.
Modern AI cameras have become flexible IoT devices. And like our smartphones, we can now think of them as platforms for host- ing specific applications for the unique job required. What will the product designers and engineers think up next?
How about AI cameras that can be trained to recognize cus- tom objects on site? It is even possible for an AI camera to ana-
lyze the video of non-AI network cameras, effectively turning them into “new” AI cameras as well.
CUSTOMIZABLE AI ON-SITE LEARNING
While AI-based security cameras have been able to significantly reduce errors by reliably detecting humans and vehicles for some time, this next phase of AI is irresistible to data-hungry business- es. Customizable AI on-site learning enables integrators and end- users to train a camera’s AI analytics on-site to recognize unique objects that are important for a business to track or count—pre- cisely what so many customers have asked for.
On-site AI training can further enhance accuracy by recognizing logos on vehicles or uniforms, counting planes, forklifts, baby strollers or shopping carts. This new stream of business intelligence data, harvested directly using edge processing within security cameras, enables more automated workflows while increasing operational efficiency and enhancing service quality.
For example, operators could teach the camera to count fork- lifts or shopping carts passing through the camera’s field of view to provide new metrics about operational efficiency. A hospital can count ambulances arriving at the emergency room. The cam- era can even be taught to recognize a logo on a truck and send out an alert when it arrives at the loading dock. The best AI on-site learning apps can even auto-generate multiple training images at different luminance values saving operators valuable training time while further increasing detection accuracy.
By processing custom data locally on the edge, AI on-site training removes the need to send sensitive data to the cloud for analysis. This is particularly beneficial for industries handling
12
MARCH/APRIL 2024 | SECURITY TODAY
EMERGING TECHNOLOGIES