Source: Unite.AI
Fredrik Nilsson, Vice President, Americas, Axis Communications, oversees the company’s operations across North and South America, leading teams focused on developing and delivering network-based security solutions for partners and end users. A longtime Axis executive, Nilsson has played a significant role in the security industry’s transition from analog closed-circuit television to IP-based network video. Since taking responsibility for the Americas in 2003, he has helped grow regional revenue from approximately $20 million to well over $1 billion and supported the expansion of more than 20 Axis Experience Centers across the Americas. He joined Axis in 1996, previously serving as Product Manager and Division Manager in Sweden and later as Director of Business Development, where he helped establish partnerships during the early development of IP video. Earlier in his career, he was a Product Manager at ABB, and he currently serves on the boards of MSAB and the Swedish-American Chamber of Commerce in Washington, D.C.
Axis Communications is a Sweden-based technology company specializing in intelligent physical security and network solutions, including network cameras, video surveillance systems, access control, intercoms, network audio, radar, and analytics. Founded in 1984 and headquartered in Lund, Sweden, Axis employs approximately 5,000 people across more than 50 countries and has been part of the Canon Group since 2015. The company introduced what it describes as the world’s first network camera in 1996 and has since expanded from network video into a broader portfolio of connected security technologies. Its current products increasingly incorporate edge-based analytics and artificial intelligence to transform video, audio, and sensor data into actionable insights for security, safety, operational efficiency, and business intelligence.
Your career has spanned remote monitoring at ABB, the early development of network-attached storage and Internet of Things technologies, and the transition from analog surveillance to network video at Axis Communications. How has that background shaped the way you think about today’s shift toward AI-powered cameras and intelligent physical infrastructure?
I see the shift toward AI-powered cameras as the next step in an evolution I’ve been a part of through my career – from connected technology and IP convergence to today’s intelligent edge. There is a common thread throughout my career, which is helping customers understand what connected technology can do for them. ABB was a great school for me. It was a large global company, so I had the opportunity to travel the world, including spending a year in Canada, and work with large enterprise customers, especially power utilities, on complex projects. My role was product manager for a new IP-connected remote monitoring and control system. At Axis, I later worked on a new category of IP-connected storage devices and helped spread the message globally about the benefits of connected technology. Then, in the early days of IP convergence, I was building on those same principles, helping enterprise customers understand the value of network-connected video and the opportunities created by IP technology. So, when I look at AI-powered cameras today, I see another step in that same evolution. The technology keeps getting more capable, but the real opportunity is still about connecting devices, generating useful data and helping customers do more with it.
What types of computer vision workloads are best suited to run directly on a camera, and which applications still benefit from being processed in a local data center or the cloud?
The workloads best suited to run directly on a camera are those where you need real-time decisions, low latency, and efficient use of bandwidth. For example, detecting and classifying people or objects, counting, tracking movement, or identifying specific events. With technologies such as Axis Object Analytics, cameras can accurately detect people, vehicles, colors, direction of travel, and other characteristics. We can also train a camera for very specific objects or scenarios through Axis Custom Analytics. As analytics continue to improve, more of that intelligence can happen at the edge. For many use cases, running the analytics on the camera and sending only the resulting data or metadata to a server, or the cloud is important for scalability and cost. The cloud and servers still have an important role, particularly for things like centralized management, long-term storage, and analyzing metadata across large numbers of devices. I don’t see it as an either-or decision. The edge and the cloud can complement each other, with each doing what it is best suited to do.
Deploying AI at the edge requires balancing model accuracy against limitations in processing power, memory, energy consumption, and heat. How are advances in specialized chips changing what can realistically be accomplished within a camera?
There are really two trends driving this. First, the AI capacity of the processor at the edge is improving very quickly. ARTPEC-9, our in-house-developed system-on-chip, has three times the AI processing capacity of ARTPEC-8, giving us significantly more room to run sophisticated models directly on the camera. Future generations will continue to increase that capability. Second, the size and effectiveness of the models are improving. That is where we have seen a lot of progress over the past year. Our R&D team has done an amazing job of getting more of the model to run directly at the edge, so in many scenarios you only need to use the data from the camera rather than sending the video somewhere else for processing. A year ago, we were talking about roughly 50% of the needs being handled in the camera. Today, it is closer to 90%, and we expect that to reach 95% or more. The real breakthrough isn’t just that we have more powerful chips. It’s that we’re combining more capable hardware with more efficient AI models, allowing an increasing amount of intelligence to happen directly where the data is generated. That makes surveillance systems much less dependent on servers and the cloud for AI.
Cameras are increasingly being positioned as operational sensors rather than solely as security devices. Which applications in retail, transportation, healthcare, manufacturing, or logistics are currently delivering the clearest business value?
Manufacturing is a very good example because the value can be tied directly to the operation. BMW Group, for example, is using network cameras as part of its AI-driven quality inspection system to capture detailed images of vehicles in real time and identify defects. That is a very different role for a camera than simply recording an incident. We are also seeing growing use of video for operational efficiency and business intelligence more broadly. In transportation and logistics, cameras can help organizations understand how people, vehicles, and goods are moving through a facility. In retail, video can provide insights into customer behavior, space utilization, and staffing. The common theme is that the camera is generating information that can be used to make a process better, not just recording what happened.
How does combining video with audio analytics, radar, access-control systems, and environmental sensors create insights that would not be possible from visual data alone?
The more context you can bring together, the better the system can understand what is happening. When you combine different inputs, you can get insights that would be difficult, and sometimes impossible, to derive from video alone. For example, you can combine what an object looks like with how it is moving, or use thermal and visual information together when lighting or visibility is challenging. Radar can add information about an object’s movement and speed, audio can identify sounds or provide two-way communication, and environmental sensors can provide additional context about what is happening in the surrounding environment. Multiple sensors can also increase confidence in what the system or user is detecting. Combining technology is all incredibly important to protect key areas of life like critical infrastructure.
We are already moving in this direction. At GSX, for example, we introduced solutions that combine visual, thermal, radar, audio, and environmental sensing. The goal is to bring together the right data so organizations can make faster, better-informed decisions. This is particularly important for applications such as protecting critical infrastructure, where having multiple sources of information can provide a more complete picture of what is happening.
Edge processing can reduce the amount of raw video that needs to be transmitted or stored, but AI-enabled cameras still raise important privacy and governance questions. What principles should organizations follow when designing systems that are useful without collecting more personal data than necessary?
The starting point should be to ask what the system actually needs to accomplish and collect only the information necessary for that purpose. Privacy and transparency are also important. People want to understand why video is being used and how their information is being handled. Our recent research shows that people are becoming more comfortable with video surveillance, but that acceptance cannot be taken for granted. The biggest concerns around AI-enabled surveillance are false positives or inaccurate information and the potential misuse or mishandling of data. That puts a real responsibility on organizations and technology providers to build privacy and security into systems from the beginning. Edge processing can help here because you can analyze information locally and send only the data that is needed, rather than moving continuous raw video around the network.
Physical infrastructure may remain deployed for many years, while AI models and cybersecurity threats evolve rapidly. How should cameras and other edge devices be designed so that their analytics remain secure, maintainable, and adaptable throughout their operational lives?
This is one of the reasons I think of the camera as part of the broader IT and operational infrastructure, rather than a standalone security device. Cyber security needs to be built into the device from the start, with secure software, regular updates, and strong cybersecurity protections throughout its lifecycle. At Axis, for example, Edge Vault provides a hardware-based foundation for protecting device integrity, securing cryptographic keys, and establishing a chain of trust. It also supports capabilities such as secure boot and signed software, helping ensure that only trusted software runs on the device.
The device also needs an architecture that can adapt as analytics and customer requirements change. Open platforms are important here because they allow organizations to integrate new applications and technologies rather than having to replace an entire system every time something evolves. At the same time, organizations need good governance around their connected devices. As more video, sensors and enterprise systems come together, cybersecurity and data protection have to be treated as part of the overall system design, not as something added later.
Axis supports an open application ecosystem that allows third-party developers to build and deploy specialized analytics on its devices. How important will open platforms and interoperability be as organizations seek to run their own models across large fleets of cameras?
I think it will be extremely important. Customers don’t want to be locked into one technology or one way of doing things, especially as AI and analytics continue to evolve so quickly. An open platform makes it easier to integrate cameras with other systems, bring in specialized applications and scale across a large deployment. It also gives customers more flexibility as their needs change. By eliminating vendor lock-in and simplifying integration, open, interoperable ecosystems make it easier to scale intelligent edge solutions while adapting to changing business demands. Ultimately, the value is not just in the camera itself. It is in how the camera, sensors, analytics, and enterprise applications work together.
How should organizations evaluate the performance of computer vision systems after deployment, particularly when lighting, camera positioning, human behavior, or operating conditions can change and potentially affect accuracy?
It all depends on the use case, but the most important thing is to test the cameras and AI models in the real environment where they will actually be used. You can have a model that performs very well in a controlled setting, but the real test is what happens with the actual lighting, camera position, weather, background, and behavior you see at the site. That is why we encourage customers to evaluate analytics such as Axis Object Analytics or Axis Custom Analytics in the real environment and make sure the accuracy is high enough for the application. The BMW factory is a good example of this approach. The cameras are being used in a real production environment, where the quality of the images and the resulting analytics have to support an actual manufacturing process.
As edge AI becomes more capable, how do you expect the role of the network camera to evolve over the next decade, and what new capabilities could emerge as cameras become active components of intelligent, responsive environments?
The network camera is becoming much more central to the system. It is no longer just a camera generating video. It is a sensor generating data that can be analyzed and shared across different systems. As the processing power continues to increase, I expect cameras to become even better at understanding what is happening in their environment and providing useful information in real time. That is the idea behind Scene Intelligence, where the camera can use a deeper understanding of the scene to provide more meaningful information about objects, their characteristics, and what is happening around them.
That could mean more advanced object and behavior detection, greater integration with other sensors and enterprise systems, and more automation around how organizations respond to events. The bigger shift is that video becomes one part of a broader data and intelligence layer. The camera is still capturing what is happening, but increasingly it can also help explain what is happening and what an organization may want to do next.
Thank you for the great interview, readers who wish to learn more should visit Axis Communications.