Sep 16, 2026
Surveillance cameras installed at distribution hubs are increasingly carrying out analytical functions that do not call for identifying a worker, including spotting an obstructed emergency exit, issuing alerts when someone steps into a danger zone, or tracing parcels along a production line, EE Times reported. That same device, though, might need to keep recognizable proof after a theft, mishap, or break-in, meaning the core difficulty for intelligent video systems lies in judging when identity is required, hiding it when it is not, and strictly governing footage that has to be kept.
Mats Thulin, who directs AI and analytics solutions at Axis Communications, told EE Times that the guiding rule is to avoid gathering or storing more material or data than a given use case demands. Axis makes professional security cameras, employs roughly 5,500 people, and works with a network of about 90,000 system integration partners, and its devices increasingly serve as real-time sensors instead of passive recorders.
Thulin outlined systems that spot people entering danger zones, flag missing hard hats, alert to pedestrians near industrial vehicles, watch for obstructed exits, and give early notice of smoke or equipment overheating. The cameras can additionally tally occupants, trace packages, and assist manufacturing quality assurance. In each of those scenarios, individuals are obscured to safeguard privacy because capturing video of people is not the goal.
Axis cameras run analysis on the spot through the company’s ARTPEC system-on-chip. Thulin said every camera in the present lineup is AI-capable, with deep-learning computation performed straight in the ARTPEC chip instead of needing separate dedicated hardware. Tools such as Axis Live Privacy Shield examine the stream within the camera, spot people, faces, or license plates, and swap out the pertinent pixels before ordinary video departs the device. Fixed zones, like windows facing private property, can be masked as well.
For an occupancy task, a camera might transmit only a tally to a dashboard. At a distribution hub, it can trace parcels while concealing employees. On-device processing also cuts the need to send nonstop video to the cloud. Thulin said that instead of sending full video to a cloud setting where it gets stored and privacy oversight is forfeited, the camera can do the counting and transmit only the needed result.
Not every camera serves only counting or process oversight. Banks, schools, retailers, airports, and transport operators may have valid grounds to keep identifiable proof after an incident. Thulin described a bank setup where ordinary users get masked video while the camera simultaneously transmits an encrypted, unmasked stream to safeguarded storage. Operator and administrator roles can keep routine monitoring apart from the power to alter privacy settings or pull identifiable footage. Thulin said that when the use case involves identifying a break-in, capturing the intruder is wanted, and attention then turns to keeping that data secure.
Axis further cryptographically signs footage so an organization can show that it originated from a specific camera and was not later modified. Long-term software support, patching, encryption, and access controls thus form part of the privacy architecture. In certain instances, Axis limits the device itself. Thulin cited acoustic sensors able to pick up screams, shouts, or shattering glass but unable to be set up to stream conversations. Privacy is stronger when needless surveillance is technically impossible rather than simply banned by policy.
Yet even tightly governed footage can turn intrusive once it is pulled out and circulated. Pimloc tackles this later phase of the video lifecycle with its Secure Redact platform. The software relies on AI to spot and follow faces, heads, bodies, license plates, scene text, and other sensitive elements throughout recorded video. An authorized user can pick which individuals stay visible, anonymize all the rest, check the results, and produce a separate file for distribution.
Simon Randall, Pimloc’s CEO, offered the case of a school sharing incident footage with parents. The parents may need to see what befell their child but ought not to get identifiable video of every other child there. Randall told EE Times that anonymization must be irreversible, and that the company essentially builds an entirely new version of the file with personal information stripped away so recovering the prior content is theoretically impossible.
Users can blur or pixelate faces or swap a whole region with black pixels. Entire bodies can be erased when a person’s stature, attire, or surroundings might betray identity. Secure Redact can additionally strip names, addresses, financial details, and individual speakers from audio, while removing metadata that might reveal time or location. The original is not always destroyed; it may stay in safeguarded storage for evidentiary needs while only the redacted copy is circulated.
Randall said anonymizing at the device is preferable and more private since the material is never captured in the first place, and that the question becomes when that method is worthwhile. Pimloc automates work that might otherwise be prohibitively slow. Randall mentioned a customer that needed three-and-a-half weeks to manually redact a notably difficult 20-minute body-camera video, and he said Pimloc cut the review stage to roughly 15 minutes.
Automation does not do away with human verification. Actual security footage can be grainy, crowded, dimly lit, or shot by a swiftly moving body-worn camera, so users have to verify that every pertinent identifier has been located and that the right people stay visible. The platform also keeps a record of what was altered. Pimloc hashes the incoming file, logs the redaction process, and hashes the finished version, a chain of custody that dovetails with Axis’s camera-level signing.
Randall said demonstrating authenticity and provenance of content is equally vital now, since privacy means not revealing private information while also ensuring that whatever is revealed is true and accurate. Robust privacy can render video more usable rather than less so. Properly anonymized footage can be shared with parents, citizens, lawyers, operational teams, researchers, or the public without needlessly identifying bystanders. Randall contended that having truly strong privacy is counterintuitive but allows far more use of the data.
Neither company regards edge AI or masking as adequate by itself. Cameras can be reconfigured, administrators can abuse access, vulnerabilities can leak streams, and a system installed for safety can steadily grow into employee monitoring or customer profiling. Cybersecurity, access control, retention limits, auditing, integrator practices, and regulation decide whether the original privacy design endures. Randall said many people set privacy and security at opposite ends of a continuum, and that both must be pursued together.
Security cameras can thus grow smarter without inevitably growing more intrusive. Privacy has to span the entire lifecycle: what the camera captures, what it analyzes, what it transmits, what it stores, who can retrieve it, and what stays visible once it is shared. The pivotal question is not merely whether AI runs at the edge, but whether the system can justify when identity matters, technically suppress it when it does not, and safeguard the original whenever it must remain.
Click Here For The Original Source.
