Step past airport security
The footstep system that aims to cut queues.
HOW YOU WALK is special– and is as good at identifying us as our fingerprints. Footstep recognition systems could replace other biometrics for security and identity confirmation, letting us keep walking instead of queuing and waiting at airports.
That’s according to Omar Costilla-Reyes, a researcher at the University of Manchester. He has built a floor sensor system and paired it with a neural network algorithm to identify people by their footsteps, which have 24 different characteristics. And it works, with an encouraging 0.7 error rate per hundred people. Here, he reveals how the system works.
Gait analysis isn’t new. What have you done differently here? This is not something we’ve created, but what we did differently is footstep analysis [rather than gait]. That means we only need two footsteps to recognise you, while most studies focus on larger gait, so you have to walk more steps, as many as six. The second difference is that we have a new algorithm that’s using neural networks, and that’s the first time it’s used for footstep recognition.
How does it measure your footstep? There are four platforms. When you start walking, you focus weight on the front side of your foot, then you complete the footstep and lift the heel. We measure the force and temporal signal from the moment you start walking until you step off; that’s your style that our algorithm analyses.
And then we can check who we think you are based on your walking characteristics. Shoes don’t matter, there’s something very special about your walk that’s not about your shoe but the style of your walk, which doesn’t change. It characterises you as a unique person.
How would this work in an airport? What passengers have to do is nothing – just walk normally. If it’s installed before the passport window, you just walk up, with the sensors installed below the floor… that capture the signals. The model needs to be trained, so the first time you go to the airport they capture the data to identify your signal. They would correlate your footstep to the identification you show, like your passport, in a database. Eventually we reach a point where we have enough data.
For future crossings through a checkpoint, it could be an extra layer of security [alongside your passport photo or fingerprint], and once it is established, they could drop other modalities. If everything goes okay, we would reach a point where you don’t have to show your fingerprint anymore, because we could use your footstep. But to do that, we need time to prove the system is resilient and works in the real world.
Why is this better than a fingerprint or other biometric? Every biometric has advantages and disadvantages – none are perfect. But the special thing about behavioural biometrics, like how we walk, versus a physical feature such as a fingerprint… this is unobtrusive. You don’t need to film anyone or make them wait; this could accelerate how people go through an airport. Imagine if you don’t have to wait in a line to see an officer to show a fingerprint or take a photo, you just walk straight.
How accurate is it in your tests? With one scenario we have almost reached 100%. We test on other scenarios, so the performance changes. The dataset we collected was very large; we had 127 users on this database of 20,000 footsteps to develop the AI algorithms based on this large data set, so we can prove it’s resilient over a large number of users.
What are you working on next? It will be interesting to mix the technologies. It won’t just be fingerprints or footsteps, it will be a more complex process. I think that will be another approach, to combine biometrics. Another future direction is to capture a longer gait, as at an airport you can capture ten footsteps or more, so we’d build a sensor that can capture more signals and see how our AI models perform.