Analysis and classification tools for automatic process of punches and kicks recognition

Abstract

This chapter deals with the pattern recognition in the time series. The data was obtained from the measurement of the force profiles via strain gauge sensor. This pattern recognition should help to classify different techniques of the professional defence (direct punch, direct and round kicks) and gender of the attacker. The aim is to find a suitable feature sets from the measured raw data which has to be transferred in appropriate way; in the case of this research spectral analysis or discrete cosine transformation were used. Based on the previous experience of authors, artificial neural networks with Levenberg-Marquardt training algorithm were selected as a classifier. In these experimentations, students from the Faculty of Applied Informatics, Tomas Bata University in Zlin participated. The results were successful and higher level than expected accuracy of 85% was achieved. The future plans include involving more participants and repeating the simulations to confirm the proposed technique.