London's railway facial recognition trial scanned 500,000 faces and produced no arrests

A six-month live facial recognition trial at London railway stations cost more than £320,000 and nearly 100 hours of police time without producing an arrest from its alerts, according to information obtained by Liberty Investigates and reported by Biometric Update.

Between February and July, British Transport Police scanned more than 500,000 faces across 18 deployments at some of London's busiest stations. The system generated one watchlist alert, which proved to be a false match.

Extended regardless

In August the force extended the trial into London Underground stations, starting at Victoria and rotating between Underground and Network Rail stations until November. "This trial is an important step in understanding how this technology can be used on the London Underground in a targeted, proportionate way with strict safeguards in place," said Siwan Hayward, director of security, policing and enforcement at Transport for London.

Images from temporary camera zones on concourses go to NEC's NeoFace M40 algorithm and are checked against a watchlist of people wanted by police or courts. An officer compares any match before acting. Images of people not on the watchlist are deleted immediately and alert images within 24 hours; the force says it gives notice before deployments, uses signs and offers alternative routes.

Measuring what a trial proves

The same algorithm produced very different results in Western Australia, where police reported 900,000 scans, 209 alerts and 79 arrests in three months. The contrast suggests results depend on where cameras go and who is on the watchlist as much as on the software.

For any public body testing surveillance technology, the London figures make a case for setting success criteria before a trial starts and publishing results against them, so that extension follows evidence rather than momentum.