Enter the maze

Shake off that foul mood

Slumped Walk

What does the way you your walk say about your mood? Maybe, quite a lot if you are wearing a neat little gadget called a SHAKE.

A whole bunch of researchers are looking at ways that computers can automatically determine things about people from the way they walk. Are they male or female? How could they improve in sports like athletics? Who are they? These are all things that might automatically be determined from a person's walk.

Mostly it's done using CCTV footage or by having people walk on special pressure pads. Parisa Eslambolchilar and her Masters student Richard Byrne from Swansea University, working with Andrew Crossan of Glasgow, wondered if you could come up with a simple and cheap device for a person to wear instead. They also wondered if you could use it to tell a person's mood from the way they walk.

Richard set to work on the challenge. He created a gadget that records the movement of a person's hips and shoulders. It's just a belt round the waist with a strap over the person's shoulder - a bit like one of the reflector belts that help make cyclists visible in the dark. The belt has SHAKEs fastened to it - one on the hip and one on the shoulder. What are SHAKEs? They are very simple, if nifty, little gadgets that contain accelerometers. It's the same sort of gadget that allows an iPhone to sense when it's being shaken. An accelerometer just detects movement in different directions. They may be a simple idea but inventive computer scientists and electronic engineers are coming up with a whole bunch of novel uses for them.

Richard created some neural networks to be the brains behind the belt. Neural networks are programs that works in a similar way to our brains, made up of artificial brain cells that can send simple messages to each other. Because they are modelled on our brains they can learn from experience just as we can. Give a neural network lots of training data (say the SHAKE data from a whole series of people walking, plus the emotion they were feeling at the time), and gradually it will start to spot patterns. Maybe a person's shoulders drop when they are sad, or rise when they are happy. That ability to learn from the data is the key - it meant Richard didn't need to know what the patterns were in advance. His software can work it out. Having trained the neural network it can then be given data where the answer isn't known - in this case SHAKE data where the person's emotion needs to be determined. The neural network then matches the patterns to the ones it's learnt. If the person is walking in a 'happy' sort of way, then the neural network can announce that they are happy, and likewise for other emotions like angry or sad where it has worked out a clear pattern.

Surprisingly, Shake data from the shoulder didn't really help

Richard created two neural networks. The first was trained to recognise different people, and the second their mood. Even when trained on a dataset of only a half a dozen people, it was 70 % accurate in working out which person it was looking at. It wasn't quite as good at recognising the mood, getting it right from the three possibilities 65% of the time. That obviously needs improvement but shows that it is in theory possible to tell a person's mood from the way they walk. With more training, it may get better still. One surprising result though was that the SHAKE data from the shoulder didn't really help. The neural network was best when using just the data from a SHAKE stuck to the person's hip.

If technology like this can be perfected there are loads of applications possible in healthcare: perhaps recognising when a person who often became dangerously depressed was starting to do so. Then help could be sent, before it was too late.

With neural networks to help us to SHAKE off bad moods, the happy times may be here again as quickly as possible.