For years, enterprise surveillance has faced a problem that has little to do with the number of cameras installed.

[For more news, click here]

Organizations can now place hundreds or even thousands of cameras across malls, ports, critical infrastructure, transportation networks and public spaces. The harder problem is making sense of everything those cameras see without overwhelming the people responsible for monitoring them.

A conventional video analytics system can identify an object, detect movement or trigger an alert when several predefined conditions occur simultaneously. That approach remains useful, but it has an obvious limitation: a single frame rarely explains what is actually happening.

A person standing near a restricted area is one thing. A person entering that area, interacting with equipment and behaving in a potentially dangerous way over several seconds is something else entirely.

That distinction is at the center of Eluviant’s latest move into what it calls video intelligence.

Formerly known as IntelexVision, the company has launched Aurora Flow, a frontier video understanding model designed for live, enterprise-scale surveillance. The technology is built to analyze sequences of video in near real time, operate across multiple cameras and run fully air-gapped for environments where video cannot leave the premises.

Speaking exclusively with Tech Revolt, Rafik Lamri, Regional Director, META at Eluviant, says the technology represents a transition away from systems that primarily identify objects toward AI that can interpret context and behavior.

“For me, most vendors are still doing frame-level object detection with rules layered on top. For example, you have a person, a restricted zone and a specific time. If all three conditions are positive, you get an alert. There is also a more advanced approach using deep-learning skeleton and action recognition, where the AI identifies the joints of a human skeleton and tries to understand movement. If a person makes an unusual movement, the system can classify it as anomalous behavior.

“But there are problems with that approach. From a deployment perspective, it can be very difficult because the camera angle, field of view and other factors affect the result. It is also primarily focused on human movement. Someone can do something unusual without making an unusual movement with their body. It also does not detect other types of anomalies, such as a water leak from an air-conditioning system, smoke, fire, a fallen tree or other events that have nothing to do with the human body.”

That is the conceptual gap Aurora Flow is intended to address.

Rather than treating video as a collection of individual images, the system is designed to reason across multiple frames and interpret what is happening over time. In practical terms, that means the AI is not simply asking whether an object is present. It is assessing whether a sequence of events represents something important enough to warrant human attention.

“So Aurora Flow introduces a reasoning layer. That is the real AI layer we add on top of the existing technology. It reasons over a sequence of multiple frames from a scene. First, it understands the whole scene, and then it understands multiple frames, allowing it to understand a video in real time.

“Imagine you have cameras connected to a brain. That brain understands what is happening, the context and the reasoning behind it. It is not just looking at a skeleton or one specific object. The whole context, the entire scene and all the frames over time are processed by Aurora Flow.

“It is not judging a single frame or an object. It is judging whether a pattern of behavior or an anomaly over time warrants human attention. It also sits on top of what we call an unsupervised self-learning engine. Because every environment is different, we first have a layer of self-learning.”

That distinction becomes particularly important at enterprise scale.

A surveillance operation can generate thousands of potential events in a single day. Most of them are not incidents. They are simply deviations from normal activity that a system is designed to notice.

The challenge is separating those signals from the events that actually matter.

From 4,000 Potential Events to Seven Verified Alerts

One deployment described by Lamri provides a useful illustration.

At a shopping mall operating 200 licensed cameras with Eluviant software, the company's self-learning engine generated 4,000 potential events over a 24-hour period. Aurora Flow then evaluated those events against the specific requirements defined by the customer.

“The other day, I was looking at some statistics from a shopping mall with 200 licensed cameras running our software. Over 24 hours, the self-learning system generated 4,000 potential events before they were filtered by Aurora Flow, all in real time.

“Some of those events were irregular movements or other anomalies. The initial layer would trigger on any anomaly, even when it was not something the customer actually needed to know about.

“When the information goes to Aurora Flow, we can tell it what to look for. For example, we can ask it to identify a person in danger or a medical emergency. The customer can build a prompt and specify many different requirements. In this case, we could tell it to identify situations such as a child playing with a scooter and falling.

“Whatever the customer requires can be included in the prompt, and the system then checks for those conditions in real time, 24 hours a day, seven days a week. Out of those 4,000 potential events, only seven were confirmed by Aurora Flow. I checked those seven, and they were all positive. There were no mistakes. Most involved children playing and falling while they were playing.”

The number is striking, but the more important point is what happens to everything that does not qualify as a priority event.

“Some of the events were unusual. For example, there was a situation where a father was playing somewhat roughly with his children and they fell down. Those were the kinds of irregularities we detected.

“I also checked samples from the other 4,000 events. I did not check all of them, but I took samples from the night, afternoon and morning, looking at around 100 to 120 events, and they were all correct.

“That means the AI is genuinely dismissing anything that is not relevant to what we have instructed it to look for. It sends the customer only the events that match the requirements they have defined. Everything else can still be made available to the customer, but it can be displayed differently, perhaps in gray, while important alerts can appear in red or yellow. The customer ultimately decides how to handle them.”

This is where AI video intelligence starts to become less about surveillance itself and more about decision support.

A control room operator does not necessarily need more information. In many cases, the operator needs less information, but information that is more relevant.

That distinction could become increasingly important as the number of cameras deployed across large facilities continues to grow.

Why Air-Gapped AI Matters in the Middle East

Another defining feature of Aurora Flow is its ability to operate fully offline.

For government agencies, defense organizations and critical infrastructure operators, sending sensitive video footage to a cloud environment can create regulatory, security and data-sovereignty concerns.

Lamri says those requirements are not limited to a small number of customers.

“All of our clients are demanding this. We mainly deal with government, critical infrastructure and defense, and data sovereignty is a procurement gate rather than a preference. There are also regulations that do not allow organizations to send video outside their premises.

“From a pure technology perspective, the cloud makes it easier to achieve smooth, continuous learning and development. But we have made the trade-off to support offline operation. We continually upgrade the system offline, step by step. For example, we can bring in an upgrade every three months rather than having continuous updates as we would with a cloud system.

“It is a trade-off, but respecting sovereignty, regulations and the requirements of the environment is extremely important.”

That model fits particularly well with the Middle East's rapidly expanding smart infrastructure environment.

The UAE and Saudi Arabia are investing heavily in large-scale public infrastructure, ports, entertainment districts, smart cities, transport networks and critical facilities. Those environments are also generating enormous volumes of video data.

For Eluviant, the opportunity is not limited to traditional security applications.

UAE Leads Deployment as Saudi Arabia Accelerates

Lamri says the UAE currently leads Eluviant's regional deployment activity in both number and scale, while Saudi Arabia has become a major area of expansion.

"The UAE is ahead in terms of the number of deployments and the size of deployments. It is definitely adopting AI quickly compared with other regions worldwide.

“Saudi Arabia is in second place. We have deployments across large public venues, critical infrastructure, ports, government facilities, shopping malls, city surveillance, entertainment venues, parks and large mosques. We are also present in entertainment environments.

“The expansion is mainly happening in Saudi Arabia right now, while the UAE is leading in terms of deployment and size.”

The regional opportunity is significant because video infrastructure is increasingly being viewed as an operational resource rather than simply a security layer.

In ports, video can support safety and operational awareness. In retail, it can help identify incidents and unusual behavior. In public venues, it can help operators prioritize situations requiring intervention. In industrial environments, it can identify behavior that could precede a safety incident.

That evolution mirrors a broader shift already underway in enterprise AI.

Tech Revolt previously examined how Axis Communications is bringing AI processing closer to the camera through edge AI, allowing surveillance systems to generate more useful insights without depending entirely on remote processing.

Aurora Flow approaches the problem from another direction. Its focus is not simply where the processing happens, but how much contextual understanding can be extracted from the video being processed.

AI as Decision Support, Not Decision Maker

That distinction also shapes how Eluviant approaches accountability.

Aurora Flow does not make the final operational decision. The human operator remains responsible for interpreting the alert and following the organization's established procedures.

“Look, Aurora Flow is a decision-support system. It is not decision-making. The final decision remains with the human operator. The system helps prioritize events and follows the organization's protocols.

“If a customer has a specific standard operating procedure, the system follows that SOP. From Eluviant's side, we are accountable for how the model performs, including its accuracy and false-negative and false-positive rates. We are very transparent about the system's limitations.

“Ultimately, the mission and responsibilities are defined contractually with each client depending on the deployment and use case. It always comes back to what the end user wants and how they define the requirements. We are a tool that enables that.”

That human-in-the-loop model is particularly relevant as AI systems become capable of interpreting increasingly complex behavior.

It also separates behavioral analysis from biometric identification, an important distinction in a market where surveillance, privacy and AI regulation are becoming increasingly intertwined.

Behavioral Intelligence Without Biometric Identification

Lamri says Aurora Flow does not attempt to identify individuals through biometric data.

“This regulation is mainly focused on biometric data. When you try to identify someone, that is where the regulation applies. Biometric data involves the unique characteristics that can be used to identify a person, and once you assign a name or match an identity, that becomes highly regulated.

“Aurora Flow is not in that segment. It is a behavior-pattern analyzer. It does not take biometric data or try to identify or match anyone. It can tell you that a person or group of people is behaving in a certain way, or that an air-conditioning system is leaking. There is no identification, matching or anything similar.”

That distinction will become increasingly important as organizations seek more intelligence from their existing camera infrastructure while navigating privacy expectations and regulation.

From Video Analytics to Video Intelligence

Aurora Flow is launching alongside another significant change for the company: the move from IntelexVision to Eluviant.

Lamri describes the rebrand as the culmination of a longer technological evolution rather than a marketing exercise attached to a product launch.

“It is a journey. The software development started in 2005, so it is a very mature platform that has gone through many layers. Initially, it was video analytics. Then we began adding AI capabilities such as self-learning, environmental filtering and deep learning.

“The more AI technology we added, the more we realized that we were no longer simply in video analytics. We reached a stage where we had moved beyond video analytics and into genuine video intelligence.

“To embrace that change, we decided to create a new brand identity and enter this segment as a video intelligence company. It is essentially a transition from video analytics to video intelligence.”

The change in language reflects a broader change in the market.

Video analytics traditionally focused on extracting specific information from footage. Video intelligence implies a more contextual system capable of connecting multiple observations and determining their relevance.

That evolution is particularly significant when surveillance systems operate at a scale where human monitoring alone becomes impractical.

The Next Step Is Understanding Multiple Cameras as One Environment

Eluviant's ambitions do not stop at understanding sequences within individual camera feeds.

The company's next development direction involves what Lamri calls group camera contextualization, where several cameras covering the same environment can collectively contribute to the AI's understanding of a scene.

“In the future, we will have what we call group camera contextualization. Instead of having one camera sending information to the brain, we will have a group of cameras, with that group providing a shared context.

“The brain, instead of watching only one field of view, will understand that this is a street and that the street is covered by three or four cameras. Therefore, if someone is doing something across those three or four cameras, we may be able to predict what happens next.

“This is where we are going in the near future.”

That development could prove particularly important for cities, transportation networks and large public spaces, where a single incident can unfold across multiple camera views.

A vehicle does not stop being the same vehicle simply because it moves from one camera's field of view into another. A person walking through a public space does not experience each camera as a separate environment.

The ability for AI to establish that continuity could move video intelligence closer to understanding environments rather than simply analyzing feeds.

From Detection Toward Prediction

Lamri is cautious about describing that future as predictive AI, particularly when the behavior involves complex human intent.

“Predictive is a big word, but I would say it depends on the vertical. In HSE, or health and safety, there is definitely a high level of prediction because people need to follow certain policies. If the system sees that someone is not following a safety policy and they continue, there is a clear possibility of an accident or harm. In that segment, we already have a high level of prediction before an accident happens.

“I would say the same applies to traffic surveillance, where there are established patterns and rules that everyone is expected to follow on the road. It is possible to identify poor driving behavior, a broken vehicle or another abnormal situation and send an alert before something happens.

“In those two segments, we are already at that stage. When it comes to pure human behavior, such as a terrorist attack or shoplifting, prediction is extremely difficult. We can detect those events when they happen, which is valuable, but predicting them is something we still need to work on in the future.”

That distinction may be one of the most important in the evolution of AI surveillance.

The value of video intelligence does not necessarily come from predicting every possible incident. In many environments, recognizing the development of a known safety risk early enough for a human operator to intervene could already represent a meaningful improvement.

Eluviant's trajectory suggests that the next generation of surveillance will therefore be less concerned with simply seeing everything and more concerned with understanding what deserves attention.

With more than 250 deployments across five continents and customers including Airbus, DP World, Prosegur and Vodafone, the company is entering that next phase with an established operational footprint.

The Middle East may prove particularly important to that strategy. The region combines large-scale surveillance infrastructure with rapid AI adoption, major smart city programs and increasingly sophisticated requirements around data sovereignty.

For Eluviant, Aurora Flow is ultimately an attempt to bridge those conditions: to make existing camera networks more intelligent, to reduce the volume of information operators need to process and to bring AI closer to the moment when a decision matters.

The broader significance extends beyond surveillance.

If cameras can understand sequences rather than simply detect objects, infrastructure can begin to respond to events with greater context. If multiple cameras can eventually be understood as parts of one environment, the physical world becomes easier for AI systems to interpret.

That is the larger transition underway.

The future of enterprise video surveillance may not be about installing more cameras. It may be about making the cameras organizations already have capable of understanding what they see.

Related Articles:

Agentic AI Is Finally Solving the Problem Chatbots Never Could, and the Gulf Is Where It's Being Proven

Andrew Ng's LearnVector Gets $100 Million From Coursera to Rethink How AI Teaches Adults

Saviynt Posts US$300 Million ARR as AI Identity Security Emerges as a Fast-Growing Market