Neural Networks Classification for Training of Five German Longsword Mastercuts - A Novel Application of Motion Capture: Analysis of Performance of Sword Fencing in the Historical European Martial Arts (HEMA) Domain

• Published in: 2021 IEEE 21st International Symposium on Computational Intelligence and Informatics (CINTI) pp. 000137 - 000142
• Main Authors: Klempous, Ryszard, Kluwak, Konrad, Atsushi, Ito, Gorski, Tomasz, Nikodem, Jan, Bozejko, Wojciech, Chaczko, Zenon, Borowik, Grzegorz, Rozenblit, Jerzy, Kulbacki, Marek
• Format: Conference Proceeding
• Language: English
• Published: IEEE 18.11.2021
• Subjects: 5 German Longsword Mastercuts; Electrodes; Europe; Forestry; Historical European Martial Arts (HEMA); Human motion database; Human motion lab; k-Nearest Neighbors; Kendo; Medical services; Motion analysis; Movement classification; Multi-layer perceptron; Naïve Bayes classifier; Neural networks; Neutral Networks; PCA; Random Forest; Training
• DOI: 10.1109/CINTI53070.2021.9668598

This paper discusses an application of motion capture in longsword fencing, a discipline experiencing rising popularity since the 1990s. Historical European Martial Arts alliance focuses on re-enacting the Late Middle Ages and Renaissance fighting styles. To popularize this art, novel research to automatically distinguish selected sword cutting techniques has been conducted. The fencing knowledge required for conducting this research was based on publications and consultation with experts in the field, and recordings. For this research, different movements from Masterstrikes such as Zornhau (Strike of Wrath), Schielhau (Squinting Strike), Zwerchhau (Cross Strike), Krumphau (Crooked Strike), Scheitelhau (Crown Strike) were selected. Motions performed by an adept fencer (acting expert) were used as patterns of correct strikes and compared with the movements of fencing amateurs. The main goal of this research was to measure the precision of movement while performing five different fencing strokes. Each movement was recorded with 39 unique full-body plug-in gait configurations initially designed for medical applications. During the exercise, 16 EMG electrodes configuration was used for the measurement of muscle activity.