German police forces have been using a facial recognition system at the Federal Criminal Police Office since 2008. Three other security or migration authorities are now expanding this technology with their own applications.
“It is now becoming clear how well police work would function if it were supported by technology, AI and facial recognition,” said the chairman of the German Police Union (GdP) on the occasion of the arrest of Daniela Klette, wanted, among others, for 30 years old terrorism charges concerning the militant Red Army Fraction (RAF). What the functionary failed to mention was that since 2008, the Federal Criminal Police Office (BKA) operates a facial recognition system known as GES, which is used by all state police forces, the Federal Police and customs. This allows unknown persons to be identified using photos – provided that these people are already stored with a photograph in the largest nationwide police database, Inpol.
Both the number of matches and the number of hits are increasing rapidly every year. The same applies to the number of people for whom biometric photos are available for comparison: Their number has now exceeded 5 million for the first time, writes the Federal Ministry of the Interior in its answer to a parlamentarian question from the former left-wing BSW Group in the Bundestag. According to this, the BKA stores around 7.3 million images of 5.1 million people in the GES – an increase of around ten per cent compared to the previous year. Due to prescribed deletion deadlines, around 358,000 photos were removed from the file, but almost one million photos were added in the same period.
The facial images in the BKA file originate from identification procedures conducted either by the police or immigration authorities. Around 3.1 million of these ED treatments with facial images were carried out by the police on suspicion of a criminal offence or other offences, while around 4.2 million originate from non-police authorities, for example when an asylum application is made.
The BKA now wants to further improve facial recognition. Having already switched to 3D technology, the Wiesbaden authority is now working on a method for artificially ageing faces. This would allow old mugshots, such as those of RAF wanted persons, in which people are depicted at a young age, to be edited and brought up to date. These aged images could then also be used to search police databases. Darmstadt University of Applied Sciences is involved in the BKA’s research project.
It was not previously known that the Federal Police also procured two systems for “semi-automated video analysis” last year. Unlike the BKA, where individual photos are compared with the Inpol file, the application is intended to facilitate video analysis of “mass data”. The state police authorities used a similar system for criminal prosecution after the G20 summit in Hamburg in 2017, processing around 3 terabytes of video and photo files.
According to the answer to the BSW enquiry, the Federal Police uses the “Investigator” software from Digivod in Meerbusch, North Rhine-Westphalia, which has already been tested in a BKA research project. According to the company, the software can also process footage from surveillance cameras and uses video analysis methods that also consist of “deep learning networks”. Investigators can search for faces, but also for people wearing certain clothing. The Federal Police want to use the two systems in “stand-alone mode”. This means that they can be quickly installed at a location where a large number of criminal offences are to be prosecuted. “Only a small and strictly limited group of people” will then be able to use the technique.
Another new development is that the Federal Office for Migration and Refugees (BAMF) is now also examining the use of new facial recognition software. The system would supplement the fingerprint comparison that has been practised for decades. In this way, the authority aims to improve the identification of people in the asylum process. Photographs from the applicant’s ID card would then be compared with the photo from the ED process in order to verify the person’s identity. A decision on a specific manufacturer has not yet been made, writes the Ministry of the Interior.
According to the description, there are no plans for the BAMF to use the new platform to search larger databases – the authority says it already has a “special software” for this purpose. In this way, the BAMF wants to ensure that the person has not already been registered under a different file number or name. With this technology, German immigration authorities are also ready for a new expansion stage of the European fingerprint database Eurodac, which will also contain facial images after a new regulation has been passed.
Finally, the Federal Office for Information Security (BSI) is also looking into facial recognition applications. This involves security aspects and the investigation of possible attempts at deception. To this end, the BSI has set up a test environment with the Biometrics Evaluation Centre at Bonn-Rhein-Sieg University of Applied Sciences, where an “AI model for facial recognition” is also being explored. There, real-life scenarios are played out using artificially generated training data. Among other things, the technology should be able to recognise altered facial features with cosmetics at passport control. To do this, the photographs in the ID card are compared with infrared images of the person in question.
According to the BSI, facial recognition systems can be tricked using other methods. These include morphing, in which the facial images of several people are fused together so that the resulting photo contains all of these features. When crossing the border, wearing glasses with specially printed frames or a special patch on a cap can also have a “significant impact” on the likelihood of matching faces, explains the BSI.
Published in German in „nd“.
Image: Authorities are testing certain measures against the tricking of facial recognition (D4FLY).
