Who needs tickets when you have a face? From today, the ticketed tourist town of Wuzhen in China is using face-recognition technology to identify people staying in its hotels and to act as their entry pass through the gates of the attraction.
The system, which is expected to process 5000 visitors a day, has been created by web giant Baidu – often referred to as the “Chinese Google”.
Wuzhen is a historic town that has been turned into a tourist attraction with museums, tours and traditional crafts. When people check in to hotels in the tourist area, they will now have their pictures taken and uploaded to a central database. If they leave and re-enter the town, the face-recognition software will check that they are actually a guest of a hotel there before allowing them back in.
Previously, multiple types of entry ticket had to be handed out to distinguish between one-off visitors and those staying for longer. But the system could easily be exploited, and some guests were caught sharing their tickets with other people to avoid paying the entry fee.
To prevent this, the town started to use fingerprint identification for hotel guests, so only one individual could use each entry pass. “But this took too long,” says Yuanqing Lin, director of the Institute of Deep Learning at Baidu.
Asking visitors to put their finger on a sensor and wait for software to verify their identity caused big queues and often resulted in false positives. The new face-recognition system uses cameras to spot people as they approach a turnstile at the entry. Faces detected by the camera are checked against a database of registered visitors, all within a second. If you’re on the database, you’re allowed in; if not, the doors remain closed.
Facing the cameras
“It was only a matter of time before face-recognition software was rolled out on this scale,” says Mark Nixon at the University of Southampton, UK. It’s more convenient to use your face than tickets, he says, so it’s likely that the technology will soon be seen elsewhere.
Baidu’s face-recognition software uses neural networks – a technique inspired by neurons in the brain that helps to recognise complex patterns. The company has trained the software on huge data sets that together total more than 1 billion images of people’s faces and says that the system has an accuracy of 99.8 per cent, although this was achieved by examining still images rather than people walking up to a camera.
The software also detects facial movements, so can’t be fooled by someone holding up a still image of another person’s face.
The system is first being used to track the 5000 people per day staying in hotels in Wuzhen, who make up around 15-20 per cent of the town’s total visitors. Baidu is already using the software for employee entry at its Beijing headquarters, but this is the first time it will be rolled out at such a scale.
Some airports already have a form of face-recognition software at passport control, but the setup is different. At an airport, you have to present your passport and the software determines whether the person standing in front of the camera matches that identity. But at the gates of Wuzhen, no identification is presented: instead, the software searches a large database for the face staring into the camera.
Compiling a database of faces in this way presents privacy concerns. Lin says the responsibility for storing the data falls to the Wuzhen attraction that uses it, not Baidu.