CCTV, Analytics and the London Underground

Closed-circuit television (CCTV) cameras have become the most widespread form of surveillance in Britain. Their output usually provides the first pictures of criminal suspects, such as with July 21's failed attempts to bomb London's public transport system.

However, with new software techniques such as facial recognition, is it possible for the authorities to use these CCTV networks to track people's movements automatically?

"In highly controlled circumstances, with a very small database of faces, [facial recognition] looks promising," says Peter Sommer, a senior research fellow at the London School of Economics. This makes it a realistic - if expensive - way of controlling access to an office, but the wider applications could still be some time away. "When you can't control the angle and the lighting, and when you're looking at databases of thousands of people, the statistics fall away," he adds.

"You have to get a person to pose in front of a camera for a given time," says Mick Napier, CCTV manager of the British Transport Police (BTP) . "Most of them have a very high false alarm rate, and are as yet not quite up to being field-proven or useful."

The need for close-ups means facial recognition is more likely to work at slow-moving locations, such as passport control or customs. "In a mass transport situation, it's still hit and miss," says Napier. BTP tested facial recognition software at Leicester Square underground station a couple of years ago, using cameras near the top of escalators to get close-ups. But even then, "it just wasn't that good," he says, especially as it required large amounts of equipment.

There are about 6,000 cameras on London's Underground, a number which will double by 2010. A Transport for London (TfL) spokeswoman says software analysis of camera output was tested two years ago at Liverpool Street underground station, to look for unattended bags.

"It didn't work on the Underground," she says. "CCTV is monitored by station control assistants, who spotted things faster than the software. The control rooms are very good at spotting things which are untoward, even a local ticket tout. That's what they are trained to do."

Dr Sergio Velastin is a reader at Kingston University's digital imaging research centre, and is also a director of Ipsotek, a firm designed to commercialise CCTV image analysis technology. Through the university, he was involved with both the Liverpool Street and Leicester Square trials.

Although unwilling to discuss those trials, he says that movement analysis software has improved recently. "It's not so much doing things humans cannot," he says. "The role of the computer here is to deal with the high volume of visual information, which we don't know if it's relevant or not, pick things up and say [to an operator], you make the decision."

Velastin says that movement analysis is well-suited to an underground station platform, as people and objects have no reason to be present for more than a few minutes, so software can compare what is stationary to an image of the empty platform. "If someone leaves something behind, we can pick that up," he says.

The same technique could be used in closed areas where any movement is suspicious, and in spotting people moving in the wrong direction on escalators or in one-way corridors. "Context is king," says Velastin.

Neither kind of software is in use on Network Rail's newly-upgraded 2,500-strong CCTV system covering its 10 mainline stations in London, which also takes input from a further 500 cameras owned by train operating companies. Last month, Siemens Electronic Security (SES) completed a pounds 17m upgrade of the system, installing new cameras and overhauling the technical infrastructure.

Camera output is stored at each station, where it can be viewed live or after an event with no interruption to recording, and is also sent to a BTP-operated facility with about one petabyte of storage (1,00GB), using one gigabit fibre-optic links, and is retained for 31 days in both locations.

Jim Kennedy, project manager for SES, says facial recognition and predictive movement technology are still fairly new: "When emerging technologies first appear, they tend to be unreliable and expensive."

Network Rail and SES are considering the use of automatic number-plate recognition technology for cameras monitoring taxi ranks and access roads to stations.