Of the 2,500 triggers, only 450 led to arrests, but the police still consider this a “loud success”.
In April 2017, the Welsh police began using the face recognition system to check out those who come to major events. The program was trained on 500 thousand images of suspects, but most of the nearly 2.5 thousand positives turned out to be false.
Official police data on the exact results vary, but in any case leave much to be desired. In the report, the police said that they received 2470 warnings from the computer, but of them only 173 coincidences were correct, and in a press release the police noted that they received 2,000 signals, 450 of which resulted in arrests.
The police explained that no recognition system “can not be 100% accurate” and noted that technical problems “will continue for the foreseeable future.” Also, police reported that part of the false positives was the result of images of poor quality sent by other services.
The face recognition system works in Wales in real time, using CCTV cameras installed in special locations or on machines. She compares the faces to a database of 500,000 images and if it finds a match, then the officer can either ignore it, or send colleagues to talk with the person indicated by the system. If it turns out that the computer was right, then the suspect is explained what happened and is invited to the equipment, where they show the document about “fair processing”.