Boxing motions classification through combining fuzzy gaussian inference with a context-aware rule-based system

This paper continues to explore the potential of newly introduced Fuzzy Gaussian Inference (FGI). It aims at constructing fuzzy membership functions by modelling hidden probability distributions underlying human motions. A fuzzy rule-based system has been employed to assist boxing motion classification from natural human Motion Capture data. In this experiment, FGI alone is able to recognise seven different boxing stances simultaneously with an accuracy superior to a GMM-based classifier. Results indicate that adding a Fuzzy Inference Engine on top of FGI improves the accuracy of the classifier in a consistent way.