Half a million faces scanned. Eighteen deployments. Almost 100 hours of police time. The bill came to £320,786. The result was one match, and it was wrong.
That is the tally of the British Transport Police trial of live facial recognition cameras in London’s busiest railway stations, which ran from February to July this year. A freedom of information document obtained by Liberty Investigates and shared with the Guardian shows the single watchlist alert during the six-month trial turned out to be a false positive. No arrests followed.
Live facial recognition, or LFR, compares video of passers-by against a police watchlist in real time. The BTP pitched the trial as a way to catch offenders and people breaching court orders at stations chosen for footfall. Supporters say the technology is exactly right for such crowds, where a wanted person can slip through a hundred ordinary commuters an hour.
The trial’s defenders point to what happened after the period covered by the FOI response. Last month the BTP announced an extension of four months, spreading deployments to London Underground stations. Since the extension, it says there have been three positive confirmed alerts of people subsequently found to be complying with sexual harm prevention orders or other court conditions. Whether that counts as success depends on what you think the cameras should be doing, and it is a fair question, because the case for LFR has always been about proportion rather than volume.
Fraser Sampson, the former biometrics and surveillance camera commissioner who now serves as a non-executive director at Facewatch, a firm that runs facial recognition in shops, told the Guardian that many variables have to align, from the choice of location to the makeup of the watchlist. He noted the difference between supermarket use, where success means a known troublemaker decides not to come in, and police use on a transport network, where success means somebody is actually caught.
There is a data protection angle that will interest anyone deploying monitoring tech commercially. Scanning hundreds of thousands of uninvolved faces to chase a handful of watchlisted people is the criticism privacy campaigners have made since LFR first appeared, and this trial’s numbers feed it directly. The Information Commissioner has so far tolerated the police pilots under strict conditions, without blessing the wider use.
None of this settles whether LFR works. It shows how hard the technology’s arithmetic is in open public spaces, and why extensions like the Underground rollout are watched so closely by companies weighing similar systems for stations, arenas and retail floors.