Milestone Systems has been at the forefront of VMS development since the days when most CCTV systems were analogue and the software backend for IP security cameras was a niche product.
Based on an open platform approach that Milestone adopted from the beginning, the Copenhagen-based company has over more than 30 years built its reputation as a VMS provider that enables end users to turn video streams into structured insights—and in times of AI, the possibilities have vastly expanded. It is no longer just about gaining basic insights, but connecting data far beyond the security sphere, in order to create intelligent infrastructures such as smart cities and others. Making big projects work, however, requires more than just AI in every connected device.
“There’s a time to be the first mover, and there’s a time to innovate at the appropriate pace and do things right. The latter is the case with AI,” said Line Hjardemaal, VP for Western Europe, Milestone Systems, as she and her colleague Martin Stockfleth Larsen, CMO, Milestone Systems, sat for an exclusive conversation with asmag during Security Essen.
asmag: What is your take on the state of the security industry right now? What impact does new technology, such as AI, have at the moment and how is that reflected in Milestone’s approach?
Line Hjardemaal: AI is moving from something people talk about enthusiastically to something organizations implement across the board. The focus has shifted to usability and the consequences of implementing AI. The sentiment remains generally positive, as organizations see greater operational efficiency, faster workflows and systems that enable better judgement. At the same time, organizations look at AI with a critical eye and ask themselves: ‘What are we basing our judgement on? Who owns the data? What is the quality of the data that the AI has been trained on? Where does the training data come from and what are their biases?’
We also see increasing legislation to govern the use of AI, such as the AI Act in the EU. Over the past 12 months, the complexity of the landscape has increased significantly. There’s a time to be the first mover, and there’s a time to innovate at the appropriate pace and do things right to get the most of it. The latter is the case with AI at the moment.
We've been in the business for decades and have been integrating AI in our platform for many years, but at the same time we're looking closely at compliance and regulations, and are building with a privacy-first principle. We want to be trusted, or rather we already have that trust with our partners and clients and we want to keep it. If there's one thing you can lose quickly, it's trust.
Martin Stockfleth Larsen: It’s also important to have an overview of the whole AI landscape in security today. AI-enabled video search and using AI to minimize false alarms is just the surface level and relatively easy to implement in compliance with the respective regulations. However, we also need to look at how AI models are being trained. We have introduced Project Hafnia, a developer-first platform through which we have access to training footage together with customers and partners under a very strict framework to protect privacy.
asmag: How does Milestone ensure your AI is trained in a way that it delivers reliable results?
Line Hjardemaal: Using high-quality, real-world, clean training data is key to developing reliable AI tools. Many of the widely available AI models have been trained on data available on the Internet, but we work closely with our partners and their own data, including a number of smart city projects that participate under the strict privacy rules Martin described. All data is being annotated and fully anonymized – using the anonymization technology from brighter AI, a Berlin-based company, that Milestone acquired last year. This means that the data is fully compliant with regulation such as GDPR. And we work with Nvidia on creating reliable AI based on this.
Martin Stockfleth Larsen: Another thing that is important to mention is biases. We sign an agreement with our partners that they cannot add biases to their data during training or build models that amplify biases, in order to make faster decisions, for example.
The rules and laws are different in different parts of the world, and we need to accommodate that. In Europe, people will think very carefully about privacy. In other regions, we work with partners that chose us in part because they like our approach to privacy. A few years back, we also started identifying every single end customer. We operate through a two-tier model, so the end customer may be quite far from us, but we want to know where our product is installed—to develop the best-possible solutions going forward but also to ensure they are used in the right way.
asmag: Across regions, in which verticals do you see the most demand for such a clean, reliable approach to AI, and maybe even great potential for Milestone because of it?
Line Hjardemaal: At Milestone, we work a lot with smart city projects. The key value we can deliver is that we use infrastructures of cameras and sensors that already exist in those cities, map out the data in the right way and help stakeholders make smarter decisions. It begins with reducing crime and making cities safer for citizens, but you can also use our platform to improve traffic flow, reduce pollution and enable better city planning.
Martin Stockfleth Larsen: In Europe and elsewhere, there’s a strong focus on reducing CO2 emissions. Sensor and camera infrastructures can identify patterns of emissions and AI can analyze them and inform how changes should be made in the cityscape to improve the situation. This can have measurable effects on the microclimate of a place. Let’s look at the example of a traffic jam for example, where many vehicles accelerate and decelerate again and again. This produces CO2, but also heat that affects the immediate environment, regardless of whether most vehicles are electric or have internal combustion engines.
Line Hjardemaal: Not all pollution peak times are plain to see to the human eye. AI, however, can identify them. It can also model ways to resolve them—through temporarily changing traffic flow by blocking or opening certain streets, or adjusting traffic light phases dynamically. Previously, cities had to count cars, pedestrians and bicycles, and make changes based on informed guesses on a trial-and-error basis. With AI, the same can be done much more precisely, and cities can model solutions that are often overlooked, such as opening certain streets only for bicycles. You can also model temperature effects of adding more tree coverage to certain streets. You can model changes in the wind pattern and much more. It’s all about making spaces more livable.
In smart cities and beyond, the challenge is how can we move from systems that respond and react to systems that also foresee and help make the right decisions.
asmag: Can you give our readers an example of how that would look like?
Martin Stockfleth Larsen: A good example comes from the US, where law enforcement uses our video analytics in a part of a city with a lot of crime. They were able to spot patterns, such as excessive foot traffic into a certain house and cars parking in front of it each for a few minutes at a time – over the course of several days. They could connect that to illegal activities in that house, which in turn was fed back into the system as a pattern of deviation that suggests crime is happening, as opposed to normal variations.
The example also shows where the responsibility lies. It's not the technology that decides to take action. Instead the AI gave law enforcement insights, but it was up to them to act. Their reaction has to be in accordance with the law, and this begins on the level of the security system. The system, too, has to operate on the basis of what’s lawful in the respective jurisdiction.
Privacy also plays a big role. In the US, students, for example, are well protected in this regard. If there is an incident on campus and video footage is obtained as forensic evidence, the system must ensure only the relevant people are being shown. Regulations differ from state to state, and systems must be designed accordingly.
Line Hjardemaal: Another example from education is vape detection, which is being installed in school bathrooms increasingly often. Obviously, the systems are not camera-based, but they can be integrated with camera systems in the hallways. When parents come in to talk with teachers about their children, they don't want to show them footage with all the other children in it. You can blur them, so parents see their own child go in and out of the bathroom, while other children in the frame cannot be identified.
This is also possible thanks to the technology of brighter AI.
By the way, this is also a great example of how strict privacy requirements spark innovation within and beyond the jurisdiction where they are in place. The initial idea behind brighter AI’s technology was compliance with EU rules, but use cases go far beyond that, from US schools to AI training data with our partners.