As The Police Are Finding Out, Facial Recognition Isn’t Very Good Yet – 80% Failure Rate

This brave new world where we are all surveyed from remote cameras is taking longer to arrive than many thought it would. The reason should be obvious enough. We humans are intensely interested in being able to recognise other humans beings. Maybe they’re someone about to attack us, maybe they’re about to offer us food, or sex, or summat. We’re really very interested indeed and we’re descended from hundreds of thousands of years worth of those who were also very interested. Thus we’re good at it.

Machines not so much:

Facial recognition technology used by Scotland Yard is wrong in the vast majority of cases and probably illegal, according to the first independent analysis of the system. Scotland Yard has been trialling Live Facial Recognition technology, in which cameras scan the faces of members of the public to compare them with faces on a list of wanted individuals. However, researchers from the University of Essex who were given access to six of ten trials in Soho, Romford and at the Westfield shopping centre in east London, found that the technology picked out faces that were not on a wanted list in 80 per cent of cases.  

It could be that not all scrotes are in fact on a wanted list but to insist upon that would be to fall into flippancy. The correct answer is just that we’re still in the early days of this technology. Google’s image search, as is well known, has great difficulty in distinguishing between a darker skin and one of our ape evolutionary cousins. This is not a great advertisement for the detailed accuracy of facial recognition:

The force maintains its technology only makes a mistake in one in 1,000 cases – but it uses a different measurement to arrive at this conclusion. The report, exclusively revealed by Sky News and The Guardian, raises “significant concerns” about Scotland Yard’s use of the technology, and calls for the facial recognition programme to be halted.

That’s really very silly as an insistence though. As long as we know how inaccurate the system is we can live with it. Anything that’s wrong 80% of the time isn’t going to get used in court as evidence – not with anything like a decent lawyer it’s not anyway.

To halt it all now would though, stop all development. It would be like stopping cars when Daimler himself was just polishing his first bumper. Technologies just do take time to mature into actually being useful.

The Neoface system used by the Met and South Wales police is supplied by Japanese company NEC, which markets the same technology to retailers and casinos to spot regular customers, and to stadium and concert operators to scan crowds for “potential troublemakers”. Scotland Yard insisted its deployments were legal and successful in identifying wanted offenders, and that the public would expect it to trial emerging technology.

Yes, quite, it’s got to be tried out otherwise it never will get any better.

There is a different area where we really should have some concern though. The systems used to check passport photos aren’t all that much better, despite being in rather more controlled circumstances. We should, of course, still be using so as to experiment with such systems but relying on them is a bit premature.

6
Leave a Reply

avatar
4 Comment threads
2 Thread replies
0 Followers
 
Most reacted comment
Hottest comment thread
6 Comment authors
Dodgy GeezerQuentin VoleQ46swannypolGrendel Recent comment authors
  Subscribe  
newest oldest most voted
Notify of
Matt Ryan
Guest
Matt Ryan

Are the ePassport gates using facial recognition? I thought the compared your face to one encoded electronically in your passport.

swannypol
Guest
swannypol

Hah. No there’s a guy in a booth. (or a gal).

The automation just gets to “Valid passport” then presents the passport photo and captured live image for a passport control operator to approve. It is more efficient as one operator can therefore cover half a dozen or so e-gates.

Grendel
Guest
Grendel

We’re very careful here to present on economic terms, and be balanced in our use of statistics. Here, we can simply see whether this technology reduces the time needed to be spent by human operators tracking persons of interest. If the technology (as I expect) reduces 1,000 hours of video footage viewed by a team of “super-recognisers” to a few hours of footage viewed by a standard operator to achieve the same overall success rate, then it will be hugely more productive even in its current form. And, as noted, it will only get better. So, the only really important… Read more »

Q46
Guest
Q46

The reason why it fails is apparent to anyone who has set up facial ID on an iPhone. To do so requires a number of scans, moving the head through different angles so information can be recorded of all aspects of the features and from different angles. I suppose it is near enough a 3D reconstruction of the face and surrounding part of the head. Interestingly, although I set mine up without my glasses, it can still ID me with glasses, even dark glasses. A photo of a subject, is 2D and shows only one perspective. The iPhone does not… Read more »

Quentin Vole
Guest
Quentin Vole

There are two different types of facial recognition. One is like the ePassport or other types of biometric authentication, where you’re effectively saying: “I am Quentin Vole, please verify”. Such systems usually require you stand in a specific relation to the camera(s), remain still, be well-lit etc. and they can work with a high degree of success (>>90%). The other type (which the article is about) has a list of targets and tries to pick them out of a crowd or scan historical CCTV footage in search of them. Such systems have the opposite characteristics in terms of image quality,… Read more »

Dodgy Geezer
Guest
Dodgy Geezer

IF the police simply extended the ‘Wanted’ list to cover 80% of the population, then the technology would improve greatly.

Thank you. I’m available for consultancy here all week. Cost to the Met – £1k per day…