Thank you, SRF!

The technological upheavals of recent years have led many people to fear for their freedom, security and independence. In fact, many modern inventions create uncertainty, and we are delighted that Swiss Radio & Television addressed one of these issues in its report about automatic facial recognition.

SRF coverage

On several channels (radio, article, podcast), SRF covered Brøndby FC’s use of cameras and facial recognition software to recognise fans banned from stadiums and deny them access. The report also mentioned the unsuccessful use of facial recognition at the 2017 Champions League Final in Cardiff, when false positives (perfectly legitimate supporters that the system wrongly classified as unauthorised) were far too high.

It was clear then that the quality of video recordings is paramount: if the recording conditions are better (exposure, view of face, etc.), the error rate decreases. According to the Federal Data Protection Commissioner (EDÖB), facial recognition software is generally permitted in sports stadiums.

SRF emphasized that there are open legal questions and that some people have concerns about whether automatic facial recognition is advisable, considering the observed error rates.
And while SRF's report highlights the opportunities and challenges in using facial recognition in sports stadiums, we would like to add a few points:
Positive Tendency
In the Champions League Final example, SRF only explored the beginning of the story. In fact, in South Wales the experiment was not abandoned, but followed up with perseverance, as the official figures and the corresponding webpage show, that even in this negative example, there is an improvement in the error rate. Detailed evaluations by the National Institute of Standards and Technology in the USA confirm this development worldwide.
Note: Evaluation only with occasions of more than 10 identifications (lower is better)

What is new in data protection law?
It is important to distinguish between facial recognition and data recording. Data is already collected on a large scale within the existing legal framework. This is the case, for example, with conventional video surveillance or the recording of personal data in the case of offending fans. The data protection issues in connection with automated facial recognition, however, concern other data: the creation of correlations among the data. In the context of stadium access, people are identified in videos, i.e. personal data is linked to video data. If these reference data are deleted after the verification process, it is difficult to identify a new data protection problem. The relevant data is already being collected.

Perspectives for Switzerland
From an economic point of view, AI offers enormous potential worldwide. If Switzerland hesitates in this development, it will destroy the foundation for technical expertise, innovation, and ultimately Swiss industry.

For industry experts, it is generally clear that the error rate in facial recognition is constantly decreasing, and the legal framework will also be staked out for very cautious companies in the medium term. Particularly in the case of security-relevant operations, such as the prevention of escalation and violence in sports stadiums, it is only a matter of time before the access checkpoints are universally supported by facial recognition. The speed with which the technology is accepted plays a central role here. In forward-thinking countries, there is a willingness to invest in technology, which gives national companies a better chance in the global market.

Fairer, cheaper, faster
It should be made very clear that visitors who have been wrongly assessed (false positives) will not simply be banned from the stadium. Access control is a process which uses automated facial recognition in order to alert humans to a potential breach. Only if a human confirms the breach is a person denied access. The introduction of face recognition leads to an increase in the efficiency of the stadium operator, improved security and less waiting time in the queue.

Deep Impact and Face Recognition

At Deep Impact, we have been researching facial recognition for years. Internal measurements regularly confirm that our algorithms are among the best in the world. For example, we measure an accuracy of 99.82% in the well-known Labelled Faces in the Wild benchmark (Unrestricted, Labelled Outside Data). To achieve this, we are constantly optimizing our software and can scale this accuracy to a very high level (i.e. process a large number of queries simultaneously).

What was initially a large investment is now paying off. From early proof of concepts to medium-sized projects, we are now developing software with a six-figure order volume, with capabilities that are only possible through artificial intelligence. Our clients include government institutions and private-sector companies. We operate in both the financial sector (RegTech) and the security industry (access control & surveillance).

AVA X and Sentinel stress test in Turkey
We have tested our automated face recognition at high-risk football matches in the Turkish national league. We achieved optimal results, returning not a single false positive.

Our Ava-X software during stream monitoring
We know artificial intelligence at Deep Impact from the get-go. Since the beginning, we have been aware of and discussed the public’s fears of facial recognition, and we can say unequivocally: AI is not a horror scenario. It is solely a collection of mathematical tools that add immense value to countless applications in business and private life.
If we use the potential as a competitive country with its very high level of education, our Swiss companies will successfully get a foothold in this industry and be at the forefront of the global AI boom.

Are you intrigued by the topic? Get in touch with us!


  • Blog
  • AVA X
  • Face Recognition
  • Venture