Bayesian reasoning for Laban Movement Analysis used in human-machine interaction

We present the implementation of computational Laban Movement Analysis (LMA) for human-machine interaction using Bayesian reasoning. The research field of computational human movement analysis is lacking a general underlying modelling language, i.e., how to map the features into symbols. With such a semantic descriptor, the recognition problem can be posed as a problem to recognise a sequence of symbols taken from an alphabet consisting of motion-entities. LMA has been proven successful in areas where humans are observing other humans' movements. LMA provides a model for observation and description and a notational system (Labanotation). To implement LMA in a computer, we have chosen a Bayesian approach. The framework allows us to model the process, learn the dependencies between features and symbols and to perform online classification using LMA-labels. We have chosen the application 'social robots' to demonstrate the feasibility of our solution.