Fuzzy C-means and hidden Markov models in modelling gesture recognition for human-machine interaction

Human Robot Interaction is an important field of research. The easier it is for people to use and co-exist with robots, the more useful robots can be for humans. Gesture recognition plays an important role in Human Robot Interaction. We examine the use of Fuzzy-C means to model posture groups and Hidden Markov Models to model the temporal structure of gestures. Gestures performed by four different people will be captured using simple infrared sensors onboard a Khepera robot. We will propose a behavioural model for the robot response to the human. Recognition rates of 98.1% were achieved for the training data set and 95.9% for the testing set when classifying between "Caress" and "Direct" gestures categories. These rates reduce to 93.3% for the training data set and 90.6% for the testing data set when attempting to discriminate between 8 individual gestures within these categories. Two other categories of gestures were also classified: "Attack", and "Play". The recognition rates between all 4 categories are 90.3% for training data and 89.3% for testing data. These rates reduce to 88.7% for training data and 86.7% for testing data when attempting to discriminate between all 15 individual gestures from the data set.