What you should know about facial recognition technology
It’s not uncommon for a person to leave their house without their loyalty cards. It is also not unheard of for person to leave home without their phone. But no one has yet to leave their home without their face.
As a personal identifier, the human face meets the key requirements of being almost unique and easily accessible. Hence facial recognition has been widely adopted as a tool for surveillance and verification purposes.
Facial recognition via computer is also one activity where computation technology is beginning to catch up with human ability. While people have traditionally outperformed machines in facial recognition tasks, especially when the subjects are moving, our ability to remember faces and names breaks down quickly as the number of subjects grows, and that knowledge is not transferrable from one personal to another.
Analytical technology, however, is now reaching levels of accuracy well above 80 per cent, even in environments where the targets for recognition are in motion, and can be applied at multiple locations and at a scale that no human can hope to match.
Geoff Cropley learnt of these limitations first-hand when he purchased a café four years ago.
“I was standing behind the counter taking people’s money for their takeaway coffee, and within weeks I could not remember the names of my customers due to the sheer volume,” Cropley tells CMO. “I could remember a face, but I couldn’t remember the name.”
Being an inventor at heart, when Cropley failed to find an existing technology solution to help him, he set about creating one himself. Cropley licenced a facial recognition algorithm and worked with a business partner to implement it as a high-speed iPad-based facial recognition system.
“We built the product for the café market thinking we were going to sell squillions,” he says. “Interestingly, everyone loved it, but no one bought it.”
That didn’t stop Cropley from developing the technology further, and today his company, Noahface, is selling iPad-based systems for access control to companies around Australia.
While even the modest cost of the Noahface self-contained system was enough to deter café owners, those costs rise dramatically as the technology is scaled up to larger implementations. The Noahface solution also did not have to contend with the increased complexity that comes from trying to recognise people in motion, as is the case in larger environments such as shopping centres and airports.
But according to the APAC general manager of Accenture-owned design and innovation firm Fjord, Bronwyn van der Merwe, the price of both digital cameras and the artificial intelligence systems used to power recognition is coming down. That is making wider scale adoption of facial recognition becoming possible.
“We are starting to see computers can see and understand the world around them around them like they have never been able to before,” van der Merwe says. “So we are seeing lots of interesting applications of this technology enabling businesses to create frictionless and quite magical experiences as a result.
“The technology is receding into the background and the experience becomes much more frictionless.”
Early versus mainstream adoption
While limited applications of specialised computing vision have been making headlines, such as the Amazon Go automated store, Van der Merwe says the main uptake of computer vision in marketing has been in retail and other mass consumer environments, where it is being used to anonymously recognise attributes of people, such as their gender, age and emotional state.
This has become a key market for the multisensory analytics company, iOmnisicent. While its technology is often engaged for security purposes, such as detecting known criminals, left objects or anti-social behaviour, it can also be used for monitoring the length of queues or for wayfinding to help people move through complicated environments such as shopping centres and airports.
One example is Singapore’s Suntec retail and convention centre, where it is used to anonymously track people’s movement through the building.
“The use cases are many, depending on who the stakeholders are,” says iOmnisicent’s founder and CEO, Rustom Kanga. “And the same system can be used by different stakeholders, each with their own purpose.”
Rival company, Cognitec has also found significant success in security applications. These include helping retailers to identify known thieves as they come onto their premises, and preventing problem gamblers from entering casinos.
According to Cognitec vice-president for Asia-Pacific, Terry Hartmann, awareness in the commercial sector of the safety and security is spilling over into more positive use cases, such as providing services to VIPs.
“Then there is the demographic information of what patronage is like, how many are coming in to the store, what is the breakup in terms of age and sex,” he says. “That is anonymous face recognition. We don’t know who the people are, but we are getting marketing statistics and marketing information based on that.”
Hartmann says Cognitec’s systems are now able to determine sex to 90 per cent certainty, and deliver a similar result for age within ten year bands. Unlike traditional people counting solutions, which might rely on an infrared beam being broken by foot traffic, facial recognition can also be used to eliminate double counting and to suppress the faces of staff out of the results. But there is much more to the technology than just being the latest solution for people counting.
“In areas like financial services, the interest is around providing better customer service to people who come in,” Hartmann says. “Their focus is on frequent and VIP customers and providing a more personal service, where if you walk into a different branch where you are not known, people can greet you by name.”
Hartmann says improvements in video analysis technology means clients can expect to get hit rates of 80 per cent or higher against their watch lists. But while interest is high, he admits it can take some time to convince a client to buy into the technology.
“We do have an amount of work we have to do in proof-of-concept stage to assure people that the software works,” Hartmann says. “Once we have done a successful proof of concept, then the take-up is pretty good from there.”
Pushing facial recognition boundaries
Numerous researchers are also pushing the boundaries of how recognition technology can be used to improve customer experience. According to the director of the Centre for Business Analytics at Melbourne Business School, Ujwal Kayande, one of the key benefits of computer vision is its ability to provide insight into what a person is really thinking.
“Human beings find it hard to articulate their preferences,” he says. “We are either very uncertain about those preferences or we are unwilling to be clear about it, or we just don’t know them. And facial expressions allow for that detection far better than any other method can.”
He is aware of one trial at Penn State University that created an automated garment recommendations system, which examined both facial and gesture responses in video footage from people standing in front of mirrors.
“They looked at where the person was touching the garment, and then used analytics to work out what the person’s preferences were, which allowed the store to then recommend other garments to these people,” Kayande says.
“What they showed was their method to recommend garments was far more valuable in terms of people picking the clothes recommended to them than a manual method of doing that.”
While the results have been promising, Kayande believes there two key factors will hold back the technology’s adoption. The first is human emotion response.
“The reaction people might have to wrong predictions might be a lot stronger than it would be if we were just human beings making the mistake,” he says. “You are better off not predicting things rather than saying things wrong.”
The second is the sheer computational power required for such systems. However, while Kayande believes this will limit adoption in the short term, he does not believe this limitation will apply even three years from now.
“The future is about understanding the unknown needs of customers that even customers don’t know they have,” Kayande says. “With facial recognition, you can get at some of that by looking at people’s expressions.”
While the discomfort that people feel about computer-based recognition remains a key deterrent from its use today, signs are pointing to this diminishing over time. Van der Merwe says the use of facial recognition to unlock the latest generation of smartphones is going a long way to easing people’s concerns about its use in other contexts.
“Before that had launched, people would have thought it was really creepy and possibly unsecure,” she says. “But now people are happy for their phone to scan their face and provide access to personal data through facial recognition.
“We as humans very quickly adapt to things where they make our lives better, more convenient, and take thing off the ‘thinking list’, and are actually quite willing to give up a lot of data and do that ‘thing’ they may have previously deemed creepy very quickly.
“But for an organisation, trust is going to be key.”