Retail comprises vastly different businesses with highly specific security needs. From the biggest department stores to malls, multi-site chains with hundreds of locations and local mom-and-pop shops, one thing, however, is the same: Businesses want to unlock the full potential of their security systems.
Those who go further and think outside the security box will discover even more possibilities, enabled by AI cameras and other technological advancements. Having smart edge devices in place is only one piece of the puzzle, unlocking their full potential is another. Foremost among the challenges facing retailers who want to go all-in on AI is handling the data produced by smart cameras and turning gigabytes into actionable insights.
Surprisingly, it is increasingly clear that the best practice for handling data is the same for retailers across the whole spectrum of businesses, regardless of size and specific needs: Hybrid architectures with local storage at the edge, integrated with a cloud solution, often provided by a VSaaS platform, offer a vast array of AI functionalities.
AI-enabled cloud features typically include advanced event analytics such as motion and object detection, heat mapping, people counting and searchable video tied to point-of-sale data.
Even the most advanced solutions are nothing without a solid anchor in a robust storage solution at the edge. That’s where Micron’s industrial-grade microSD cards come in. Engineered to handle continuous, high-bit rate recording and preserve critical footage even when network connectivity is unstable, they enable retailers to harness the power of edge AI.
The long road to edge AI
One of the most obvious reasons retailers started installing security cameras many years ago was to deter potential shoplifters and identify suspects. At that time, staff realized items were missing. They would notify security personnel, who would search through hours of security footage for the scene in question — a painstaking task. And it wasn’t certain that they’d be able to identify a subject from the grainy footage.
Modern security cameras don’t just boast better image quality; they also offer enhanced security features. With HD recording in 4K, high dynamic range (HDR) and wide-angle lenses that capture more detail than ever, they allow operators to retrieve high-quality information from the large amounts of data they produce.
Today, AI-enabled camera systems can also filter security footage, with many offering natural-language search. Operators can ask the systems, “show me the footage of the person hiding a black sweatshirt under their coat,” and the system produces the right scene within seconds. Systems that automatically flag suspicious behavior and send event alerts are the next frontier.
Obviously, this further increases the amount of data that security cameras produce. 4K cameras streaming at 30fps typically generate between 60MB and 120MB of video data per minute, depending on codec and settings. In AI-enabled systems, however, metadata bursts can increase random write workloads up to 10-fold, when the AI “describes” what it’s seeing in the frame. Even though these bursts last only a few seconds, storage demand significantly increases, especially when considering a simple fact: The more advanced the analytics, the more frequent the burst, the bigger the amount of data produced.
Edge AI unlocks new functions
One of the most common, cloud-based AI features is facial recognition. This technology is increasingly used in retail scenarios, as businesses require cameras to detect more than just theft. Take, for example, the prevention of deceptive practices: Modern security systems can detect when a person tries to return items multiple times across multiple branches of the same chain, even though they only bought the item once.
The same technology is used to recognize VIP shoppers and offer them special services, as well as track individual items across the premises of a retail site to ensure operators have full visibility — from the parking lot to individual shops and the food court.
Systems increasingly come with high privacy standards, for example, the encoding of facial data to ensure no one can reconstruct recognizable faces in case of a data breach. This, too, increases the data storage demand.
Cameras can also flag disruptive behaviors. Once again, analyzing images and adding metadata — such as “person hiding their face” or “person hiding merchandise” — increases data storage demands.
Why edge storage is crucial
One might now ask the central question: Why does there have to be AI both at the edge and in the cloud? Couldn’t AI, centralized in either of the two places, do the job more efficiently?
The cloud seems to be the better place for advanced AI, thanks to virtually unlimited computing power and the ability to centralize intelligence. Why is edge AI — and storage — even needed? Wouldn’t it be better if edge devices streamed raw footage directly to the cloud, rather than storing so much data?
The best argument against “cloud-only” AI and storage is the bandwidth required to stream data — up to 120MB per minute plus metadata generated in bursts — to the cloud. Loss of connectivity, or even delayed transmission, would weigh heavily on the system’s performance. Real-time analytics would not be possible, only delayed analytics. Raw footage streams would also incur prohibitive bandwidth costs, especially for retailers operating multiple sites with multiple cameras at each site.
In systems where edge AI and cloud AI are fully integrated, cameras generate metadata in bursts, for example when activity is detected and needs to be classified, which increases the random write load up to 10-fold compared with non-AI cameras that only stream video. As the primary processing unit, the camera decides what data to stream to the cloud, while retaining the full volume of data at the edge.
Both the read/write and storage demand at the edge is substantial, and it is only expected to grow as AI capabilities advance further.
Aside from the best-possible distribution of AI, hybrid cloud-edge architectures offer many benefits to retailers. Large businesses and mall operators benefit from scalability and the ability to outsource system management to VSaaS providers while keeping local footage instantly accessible. Multi-site retail chains save costs by eliminating the need for separate recorders at each branch — an important advantage when multiplied across hundreds of sites. Even small, family-run businesses can enjoy advanced analytics that were once reserved for enterprise-level installations, all while maintaining plug-and-play simplicity.
Thanks to the highest quality standards guaranteed by relentless testing, industrial microSD cards by Micron boast 2 million hours mean time to failure (MTTF), with an annualized failure rate (AFR) of about 0.44%. Micron firmware is designed to minimize frame drops, with additional features such as health monitoring, helping integrators and VSaaS providers predict and prevent failures before they occur.
With capacities ranging from 64GB to 1.5TB, industrial microSD cards by Micron give integrators the flexibility to tailor hybrid systems to the needs of every retailer — including those required to retain footage locally for extended periods. With 30 days being increasingly the norm and some jurisdictions pushing for up to 90 days, industrial microSD cards by Micron are at the edge of what’s technologically possible, and offer what retailers need to make their systems future-ready.
As retailers continue to adopt AI-driven video security and analytics, the importance of reliable edge storage will only grow.
Micron’s industrial microSD cards stand out as a cornerstone for these hybrid systems — enabling faster insights, greater resilience and future scalability in an era where every frame and every metadata tag enables advanced, intelligent systems.