Gait Expectations

If you navigate Associate Professor Terence Sim’s profile on NUS Computing’s website, you’ll see a smiling, affable-looking gentleman in a suit and tie.

But mouse over the photo and it changes into him in what looks like Olympic sprinter Usain Bolt’s victory “To Di World” pose, an altogether stark contrast to the earlier professorial-looking photo.

To some, the two photos are both reflective of Assoc Prof Sim’s genial nature - that is well known among his colleagues and students at NUS Computing, where his research interests focus on, among other areas, biometrics and computational photography.

Look past the funny pose, however, and therein lies a deeper connection to running.

How can gait identify you?

As an avid runner, Assoc Prof Sim is well aware of his gait, which is a person’s manner of walking. And that got him thinking – is there a way to identify a person using their gait? The short answer, he says, is yes.

“From your gait, we can verify your identity, pretty much with more than 90 per cent accuracy,” he shared.

Surprising? Not really.

He added, “Gait is unique to an individual, whether you're walking or running, and it can be used as ‘constant authentication’ for your smartphone – or apps on your smartphone – with advantages over other forms of authentication.”

Constant authentication refers to the mobile phone “knowing” that it is with its rightful owner at all times, so it doesn’t have to constantly guard itself. If the phone is handed to someone else with a different gait – or even stolen – it detects that change in gait and can then lock itself or send out an alert.

Assoc Prof Sim’s research focuses on collecting and analysing data gleaned from an inertial measurement unit (IMU), a device on a tiny chip about the size of a grain of rice that can detect and measure how the body is oriented. The IMU combines accelerometers and gyroscopes, and can be found today in devices like smartphones, tablets and even wireless earbuds.

His research breakthrough showed that personal attributes could be inferred from gait to a reasonable degree of accuracy.

“In practical terms, any app that reads your gait signal – perhaps from your smartphone or smartwatch – knows more about you than you may wish to reveal,” Assoc Prof Sim said.

He discovered that by using a smartphone’s IMU, he could tell in just 0.3 seconds – or one-third the time it takes you to take a step – a person’s gender, their age, how much they weigh and how tall they are. His research can do all this to a significantly accurate level.

“If you compare this with baseline – which is basically random guessing – there is a huge improvement over that. Gait alone can reveal all these physical attributes,” Assoc Prof Sim said.

For those who are worried about privacy, fret not: gait cannot precisely pinpoint intensely personal areas like your personality – e.g. whether you’re an introvert or extrovert – or income level.

Among other forms of authentication, gait is actually one of the most private as it infers information about your gait from the phone’s IMU, as opposed to using a readily-identifiable physical attribute such as your voice or face.

It is also the most power-efficient way to authenticate a smartphone user. Assoc Prof Sim has studied other forms of authentication on a smartphone, including facial recognition. While facial recognition on a smartphone affords 100 per cent accuracy, it also uses significantly more battery power.

“We did energy profiling of all the biometrics available on a smartphone. Among all of them, the IMU consumes very little power, yet gives us good accuracy as a biometric. Compared to the others, gait is cheap and good!” he said with a twinkle in his eye.

Getting a head start in computers

Assoc Prof Sim got his start in biometrics nearly 20 years ago while working on his PhD in electrical and computer engineering at Carnegie Mellon University. He started working on face recognition, then branched into other biometrics like fingerprint and iris recognition.

He was interested in computers ever since he was a teenager in the 1980s, being self-taught on very early computers like the Apple II (“they were the latest machines then!”).

“A few of us were very interested in computers, so we just picked up some library books and started to learn and tried to get hold of some computers.”

Computers at the time ran at what would now be considered glacial speeds of 8 megahertz (MHz), with a tiny 48 kilobytes (Kb) of random-access memory (RAM), he said. As they were expensive, Assoc Prof Sim never owned one until he entered the workforce after graduating from university.

“Even so, my first computer cost me around $4,000 as I asked for a larger 120 megabyte (MB) hard drive,” he quipped.

Other research interests

Alongside his research into gait, Assoc Prof Sim is also actively studying and developing methods to detect fake videos, with the eventual aim of curtailing their spread.

Fake videos are manipulated videos to show someone saying things they never did. These are especially damaging if used on influential individuals like politicians or celebrities, or used during national elections. Broadcasting manipulated videos could easily deceive the masses and trigger serious consequences.

“In time to come, such videos will become more prevalent and it will be harder to distinguish from real videos. Seeing is no longer believing,” Assoc Prof Sim said.

Thankfully, widespread proliferation of such videos is not yet a reality as it currently requires professional techniques.

Given a crystal ball, how does he see technology in about five years?

“We will definitely see a lot more connectivity with our daily appliances, thanks in part to 5G and the Internet of Things. Almost everything will be connected to networks, and there will be a lot more exchange of information,” Assoc Prof Sim said.

In future, artificial intelligence will also be behind more technology as there will be more data generated. In order to crunch such large amounts of data, more automated tools will be needed.

“I tell my students, there will be jobs in cybersecurity and data science, for sure!” he summed up.