Fine classification of complex motion pattern in fencing

Abstract

The subject of this study was fencing and the object was to classify the fundamental motions of fencers by creating a library of movements. Based on this library, thus, the recognition of motions during a real fencing match can be made. Kinematic data were acquired by a motion capture system (Vicon). The automated algorithm that recognized motions is based on three steps: a Principal Component Analysis for data dimension reduction, an innovative wavelet-based analysis of signals and a feature extraction method. The algorithm was tested on high level fencing athletes and it was found to be robust with a 12% of misclassification rate. It gave a description of how atheletes move and showed that in real match athletes do not execute fundamental motions but they mix different techniques in order to surprise the opponent.