By Katie Scott, Wired UK
The way we walk could be used as an accurate way of identifying us, according to an international team of bioengineers who analyzed the foot pressure patterns created by 104 subjects. They found they were able to identify individuals with a 99.6 percent accuracy.Plantar Pressure Imaging, which is a system that consists of an array of hundreds or even thousands of pressure-sensitive sensors. The team deployed a system that could record both the placement of the foot on the sensors; and when the pressure was recorded, on a scale of about 5mm and 100 Hz, respectively. This allowed the team to record the speed with which the participants “transitioned to different parts of the foot”.
PPI has been tested in previous studies, but, says lead author Todd Pataky, an assistant professor at Shinshu University, the results were limited by a small sample size, yielding an identification rate under 90 percent.
Speaking to Wired.co.uk, he added that his team made two key improvements — normalizing each footstep so they can be directly compared, and simplifying the foot’s features: “We used a better method of (…) normalizing the data, and we used much simpler features than in previous studies. We showed that you can use very simple features if your (data) is good.” That normalization was made possible by an image alignment tool, which both aligned the images in terms of positioning within a grid and also in terms of time so each recording started at the initial heel strike
The classification rates were boosted further by using algorithms to reduce the vast amount of data the team had to sift through — with approximately 100,000 pressure values for a single step. John Goulermas, co-author of the study, explains: “Algorithms are capable of reducing these features to significantly fewer ones, without destroying the intrinsic statistical data distribution. They retain the ‘gist’ of the data and get rid of noise and extraneous or irrelevant information. This benefits not only storage but the classification algorithm also works better as it can focus more on the features that are more discriminating in terms of the classes.”
But could this system be deployed in airports to identify passengers? Pataky says that the team need to next test their technique with a larger number of subjects. Also, to date, they’ve only tested with barefoot walking, which could be impractical in an airport.
Then it would need funding: “A security company would have to become interested, assess its financial feasibility, and support continued development. The technology is not ready to use out-of-the-box, it would require integration with security databases and would also require software optimization, things that are not necessary for research. As researchers we are unable, nor are we inclined to commercialize this technology ourselves.”
On the bright side, however, Pataky says that the number of people that such a system could track would be limited only by the size of the sensor array. We just wouldn’t want to be responsible for cleaning the array after thousands of sweaty-footed passengers had passed through.
The research was published in a paper in the Journal of the Royal Society Interface.