TECH

Facial recognition of London police misses 81% of cases

The first independent assessment of the facial recognition system used by the London police shows worrying data: according to the study, facial recognition systems have an astounding 81% error rate.
The study, conducted by the University of Essex, examined all cases where the recognition system was used by police to seize people since its installation in 2016. The study found that of the 42 people who were identified by the system, only eight were even on the wanted police list. And considering that at that time 10 suspects were put on trial for being identified by these cameras, this means that at least two innocent people were forced to go through the entire criminal arrest process.
Since 2016, London police have used NEC's Neoface facial identification system which, based on the city's surveillance cameras, promises to recognize fugitive suspects and identify them to authorities. Although the University's study reveals how inefficient the system is, London police disagree with it, and claims it has evidence that the system has a margin of error of only one in 1,000 identifications - but it does not reveal to anyone the full study or even the methodology used to reach that number.
The Essex University study makes it clear that one of the main points critics of this type of recognition system is correct is that they do not yet have the necessary precision to be treated as criminal evidence. For this reason, last month the city of San Francisco banned the use of any type of facial recognition tool by its security organs. And as long as those systems are not really accurate, this should be a path taken by police departments around the world. Source: MIT Technology Review