Cameras scan the faces of shoppers in the biggest mall in Rio de Janeiro, pinging alerts to the mobile phones of shop assistants if they recognise known shoplifters.
If it spots anyone wanted for a serious crime, the system calls the police.
The facial recognition program is being rolled out across Brazil this year, including in football stadiums, other shopping centres and gyms, but it began life in Gordon’s Wine Bar, a Victorian drinking cellar on the north bank of the Thames in London.
When 62-year-old Simon Gordon inherited the bar from his father in 2003, he was struck by the rampant pickpocketing that took place in its candlelit interior.
“We had 15 crimes every month, two of my friends had their wallets stolen while I was having lunch with them,” he said. “They came in spates and the police could not deal with them, they just did not have the resources.” So, in 2010, he decided he would take action himself.
Today the company he set up, Facewatch, is being tested by a major UK supermarket chain, several major events venues and even a prison. It uses off-the-shelf facial recognition software to match criminals against police watch lists, and against watch lists compiled by its customers.
In Brazil, it has been used by three commercial centres since last May and was lauded by Wilson Witzel, the governor of Rio de Janeiro who was an early supporter of Jair Bolsonaro and echoes the president’s tough-on-crime rhetoric. Mr Witzel said in January that he plans to allow the police to share their watch lists of members of organised crime suspects with facial recognition companies.
“In December alone, we captured 2.75m faces [in Brazil] and reduced crime overall by 70 per cent. We even caught one of the top 10 criminals from Interpol’s Most Wanted list, which had been uploaded to our watch lists,” said Mr Gordon.
The first incarnation of Facewatch was a tool that allowed customers to upload footage of crimes captured on CCTV and send it to the police.
Mr Gordon, who trained as an accountant, built the system with three friends from his village in Hampshire and launched it with 50 retailers in the area around London’s Victoria railway station, including Co-op Food.
By 2016, more than 20 police forces were working with Facewatch, but cuts to their budgets meant they had few resources to investigate the incidents being uploaded. “There was no point reporting low-level crime, if the police weren’t responding,” said Mr Gordon.
But at that point, facial recognition software was starting to become widely available, thanks to advances in artificial intelligence, and Mr Gordon decided to try it out on the £3,000 camera system installed in his wine bar.
“When we first started, faces were really technically difficult to identify in real time, and very expensive to do, you needed powerful processors that most people couldn’t afford,” Mr Gordon said. “Our idea has coincided beautifully with facial recognition technology becoming market-ready, it’s 20 times faster than it was three years ago.”
In 2017, the company, which has just raised £2.5m in seed funding, brought on a new chief executive, Nick Fisher, to sell its alert system across the UK and Brazil.
Start-ups like Facewatch, and others such as US companies StopLift and FaceFirst, which supply to large retailers including Walmart and Saks Fifth Avenue, have benefited from the huge recent strides in machine learning, the technology by which computers can learn to recognise faces or objects by training on large data sets. Consumer giants such as Amazon and Microsoft sell their own facial recognition tools to developers to integrate into new apps.
Facewatch’s system uses an algorithm from US company RankOne to match faces seen in real time by in-store cameras with watch lists of known criminals.
The process takes a few seconds, and if a match is found, the business owner is notified by a mobile alert. Facewatch is about to sign data-sharing deals with the Metropolitan Police and the City of London police, and is in talks with Hampshire police and Sussex police.
“The deal with police is they give us face data of low-level criminals and they can have a separate watchlist for more serious criminals that they plug in,” said Mr Gordon. If the systems spots a serious criminal, the alert is sent directly to the police, rather than to retailers.
Facewatch plans to charge roughly £2,000 per camera for a three-year licence, but said it is in “beta” mode, with most of its customers trialling its products.
Humber prison, in north-east England, is using the system to spot prolific drug smugglers, said Mr Gordon.
“Rory Stewart, minister for prisons, allocated £10m to 10 prisons in the north and said he would resign in 12 months if he didn’t reduce drugs in those prisons. Our trial in Humber ends this month and then they’ll test it at the others. You put this in multiple prisons and you can spot the same people going round all the time,” said Mr Gordon.
Today, he does not spend much time running his wine bar, which he only visits twice a week, he said.
“We are, of course, using Facewatch there, and we now have one crime a month maximum, it’s dramatic,” he said. Stuck up behind the bar is a poster of five faces. “We have so little crime now, that we hardly have anyone left on our watchlist.”