Fuzzy Qualitative Gaussian Inference: Finding hidden Probability Distributions using Fuzzy Membership Functions

This paper introduces fuzzy qualitative Gaussian inference: a novel way to build fuzzy membership functions that map to hidden probability distributions underlying the informationally structured space. This method is used to classify boxing moves from natural human motion capture data. In our experiment, the system is able to recognise seven different boxing stances simultaneously with an accuracy superior to a GMM-based classifier. Results seem to indicate that a template can be learned and a stance identified in under 18 milliseconds, which may allow recognition in real-time.