You may enter: automated ID checks face scrutiny

Athletes at the Tokyo Olympics will be relying on computers automatically matching their face to their picture to make sure they aren’t late to the start line this summer. Tournament security will be boosted by real-time facial-recognition technology at venues where available space is too limited to accommodate manual identity checks and the queues those procedures often create.

More than 40 locations, including the main stadium, International Broadcast Centre and the Olympic Village, will be covered by the system, which is designed to ensure that over 300,000 accredited athletes, staff, media, and volunteers can be quickly verified and admitted.

It won’t replace physical checks completely, and compares a picture ID embedded on a traditional lanyard security pass against the face of the person seeking admittance, most likely with greater speed and accuracy than human staff could achieve for the same comparison.

Rather than being centred on a single, centralised Olympic Park, Tokyo’s sporting venues are spread across the city. Athletes have to be able to travel freely between these locations and authenticate themselves at each one, and the potential for delays that could impact a tight schedule of events is significant. Organisers are also concerned about the health implications of having people spend too long standing in a queue in the summer heat (Tokyo’s average July temperature is in the range of 23-30°C).

Facial recognition has another distinct advantage over other forms of biometric identification (such as fingerprint recognition) because it is contactless, doubly important to the Tokyo 2020 Organising Committee this year because it reduces the chance of coronavirus infection caused by touching of shared surfaces.

The system was previously trialled at the 2016 games in Rio de Janeiro and is based on the NeoFace Watch technology developed by Japanese IT and electronics giant NEC. NeoFace Watch uses the generalised matching face detection method, within which is an artificial neural network model called the generalised learning vector quantisation algorithm that searches and selects facial features for matching purposes.

The perturbation space method converts two-dimensional images like photographs into three dimensions – a process called morphing – so that the candidate’s head can be rotated left and right and up and down to achieve greater accuracy. Further processing applies different levels of illumination filters across the subject’s face to facilitate an even better match of the photograph against an image stored in a central database.

‘All personal data is managed and used appropriately during the games and securely deleted afterwards under strict conditions.’

Accurate facial recognition can still be hampered by varying expressions – such as a smile or inadvertent blink of the eye – as well as the wearing of head accessories such as hats or sunglasses. To combat those effects, NeoFace Watch uses an additional process called the adaptive regional blend matching method to minimise the impact of local changes.

NeoFace Watch comes highly rated by the US National Institute of Standards and Technology (NIST), which measured its ability to match face image data against large databases of people in personal identity searches as well as ‘one to one’ matching capabilities where two face images are evaluated to determine if they are the same person. Subsequent NIST testing suggests the system can check 1.6 million faces in as little as 0.3 seconds, with an estimated accuracy of 99.7 per cent.

It is not clear if the processing will prove that quick in Tokyo, or whether it will take place on the hundreds of high-definition video cameras installed around Tokyo venues that monitor the entrance facilities. More likely the number-crunching will happen on attached peripheral devices such as web-based thin-​client terminals powered by Intel processors.

The chipmaker launched its own RealSense ID system in January this year. This uses a small module with a dedicated system-on-a-chip component to securely process and encrypt user data that can be easily integrated into other products, delivered as a standalone peripheral which can be plugged into any computer.

Concerns around data privacy have inevitably risen, not least because NeoFace Watch is just one component of NEC’s Bio-Idiom line of biometric authentication technology, which has been widely embraced by criminal surveillance and other enforcement organisations around the world. It is currently in use at New York’s JFK Airport, for example, where it is used to check faces against a photo database of ‘undesirables’ and send an alert when a match is found.

Another deployment is with the New South Wales police force in Australia, which uses the system to match real-time CCTV footage from live camera feeds to a watch list of persons of interest, and to compare crime scene images against those stored in databases to identify similarities.

The Merit Lefkosa Casino in Northern Cyprus uses NeoFace Watch to track ‘problem’ gamblers’ activities and unauthorised individuals not already included on its white list, while the same system identifies VIPs to provide them with faster access into its game rooms.

Reports suggesting the system will also be used to provide contact-tracing capabilities to help monitor any spread of coronavirus during the Games remain unconfirmed, though we do know that NEC has already developed the technology to monitor people and groups for compliance with transmission reduction guidelines. NeoFace Watch Thermal, for example, combines facial-recognition and thermal-imaging technology to make venues more Covid-19 secure using specially equipped CCTV cameras, which also record the date and time of somebody’s visit to a specific location.

The Tokyo 2020 organisation has not said how long it will keep people’s personal data, but has promised to comply with local privacy law. In Japan this falls under the auspices of the Act on the Protection of Personal Information, originally passed in 2003 and amended most recently in 2020 (the government sensibly mandated in 2015 that updates should be considered every three years to keep pace with the rapid evolution of technology and global standards).

“All personal data is managed and used appropriately during the games and securely deleted afterwards under strict conditions,” the organisers say.

NeoFace Watch will only be applied to athletes, staff, media, and volunteers – those will have given prior consent when they submitted their photographs for inclusion in the central database and on the identity card alongside an embedded IC chip containing their personal ID and authentication details.

The organisers insist it will not be used on the general public or spectators who have not given the same explicit consent.

In the unlikely event that the data is not collected, stored and processed properly and legally, in accordance with Japan’s privacy laws, it may take time to be exposed. But what might be harder to conceal under the intense media spotlight are mounting queues outside venues if the accuracy of the NeoFace Watch system fails to match organisers’ expectations.

On very public trial is not only the technical prowess of the athletes themselves, but also the efficiency of AI-driven facial-recognition systems used for ID checks and queue management which may lead to extensive deployments in other scenarios long after the closing ceremony is complete.