MRes (2008-2010) - Swansea University

I Can Tell by the Way You Walk:
Mood Recognition via Gait Phase Effects

How are you feeling today? What mood are you currently in? These are questions that are trivial for us to answer ourselves but how can a computer system make sense of these emotions and detect the mood of a user? Can this be achieved in an unobtrusive and inexpensive way? This is the key question and main discussion of this thesis. Within we discuss the creation of a mobile sensor based gait recognition system which we use in order to find if we can identify the mood of individuals based on their walking information.

We describe three studies the first of which examines the best body positions to at- tach the sensors in order to gain gait information. Following this we conduct a study with seven participants whom we ask to simulate three separate moods - angry, sad and happy, and ask them to walk a route four times (once to record an initial walking pattern before they walk simulating any moods) whilst recording their gait information. We then train several Neural Networks with the gait information we gathered from the seven participants and attempt to recognise individuals and their resulting mood by submitting test patterns to the networks. We discuss how we managed to achieve an individual recognition rate of 70% and a mood recognition rate of 54%.

In our third study we detail improvements we made to the procedure of the second study and discuss a new method of managing the gathered data in an attempt to achieve a higher success rate from the Networks. We run a new study using eight participants and repeat the training procedure for new Networks using the new gait information. This time we detail how we managed to improve person recognition to 75% and mood recognition to 73%.