Blog
It’s a familiar scenario for police forces: it’s the weekend, a football match is underway, and tensions in the stadium are running high — suddenly, things escalate. Dozens of hooligans ignite pyrotechnics, rival fans are attacked and injured. Scenes like these are unfortunately not uncommon. They not only ruin the experience for law-abiding fans, but also present investigators with a significant post-incident workload. Because one thing is clear: the perpetrators must be identified and brought to justice. The digital search for clues begins.
There is often no shortage of data. Investigators can draw on footage from fixed surveillance cameras in and around the stadium, as well as publicly available videos and images shared on social media or the internet. Yet investigators face multiple challenges. First, the material must be reviewed, sorted, and linked — often involving terabytes of data. Second, the footage frequently shows individuals wearing masks or other coverings, as suspects are typically well aware of the surveillance measures in stadiums. So what then?
One software tool assisting law enforcement in such searches is the object and person recognition solution from rola Security Solutions. The software utilises neural networks and independently detects patterns. It automatically scans vast amounts of visual material, highlighting relevant objects and details for analysts to review. It doesn’t just recognise individuals, but also weapons, vehicles, and symbols. In our football stadium example, this means that even if a suspect is wearing sunglasses and a scarf over their face, other matching features can still be identified.
"This ability to match features using AI-supported technology is a genuine gamechanger for investigative work," explains Dominik Kahsche, Solution Consultant at rola. “The new ‘person-region search’ feature allows analysts to focus on distinctive traits or specific areas of the body. A stadium surveillance image may show a suspect wearing a balaclava — initially making identification impossible. However, if the person is wearing, for example, distinctive trousers, a branded logo on their chest, or even a notable tattoo on their upper arm, investigators can isolate and match these regions with other images.” This allows a tattoo to be identified on additional footage — such as of the suspect approaching or leaving the stadium — where they may not be wearing a balaclava.
The key to this level of precision lies in the transition "from pixel to vector", powered by advanced neural networks. These networks don’t just analyse individual pixels — they detect complex patterns and features across vast datasets. Central to this transformation is the use of vectors, which can precisely describe the uniqueness of characteristics such as the arrangement of the eyes, nose, and mouth. Unlike traditional pixel-based methods, vector representations are less affected by changes in lighting or perspective.
In simple terms, the neural networks are trained to generate vectors in such a way that similar faces result in similar vectors, while dissimilar faces produce distinct ones. These vectors then serve as the basis for determining whether two faces or objects are identical or different.
Think of a vector as a kind of secret code that defines a point in space. Made up of a group of numbers, it provides not only a position but also a direction. While this might sound like dry mathematics at first, vectors are particularly exciting in image and speech processing. In these fields, they help decode the fascinating details of images, texts, or audio signals, and convert them into comprehensible information.
In the world of technology, vector-based person recognition is comparable to the leap from 2D to 3D — more complex, more accurate, and truly revolutionary.
As with all technology, one principle always applies: the results provided by the software are suggestions to support investigations, not definitive identifications. It remains essential that the outputs are interpreted and evaluated in context by human experts. The technology enhances workflows and optimises processes — but never replaces the critical importance of human expertise.
Person recognition software continues to be a topic of public debate, particularly when it comes to balancing utility and risk. “And rightly so,” says rola expert Dominik Kahsche. “All too often, safety and data privacy are neglected. At rola, we regard data protection as paramount. That’s why all of our products comply with the highest standards in data privacy.”
In the world of investigative work, efficiency is key — and this is precisely where rola’s software solutions come into play. From seamless evidence collection to in-depth review and rapid report compilation, they provide investigators with optimal support. Technology serves as a partner in the investigative process, enabling security professionals to leverage their expertise even more effectively. The combination of human judgement and advanced software creates a powerful synergy for successful investigations.
Dominik Kahsche has been a Solution Consultant at rola since late 2022. With many years of experience in IT and sales, he now connects user needs with market opportunities and contributes to the ongoing development of rola’s product portfolio.