Application of Machine Learning Technology for Automatic Motion Recognition for Japanese Tea-Making Procedure of CHA-NO-YU

Abstract

CHA-NO-YU, the Japanese traditional tea-making ceremony, has been a research subject in many academic fields including anthropology, psychology, and engineering. To effectively study the CHA-NO-YU, automatic recognition of motions of the tea-making procedure is desirable. Hence, herein, a method is proposed to recognize 12 motion classes of the tea-making procedure with acceleration, angular velocity, and right-hand tilt angle data. In the experiment, one Japanese subject with 18 years of experience in CHA-NO-YU performed the 12 motion classes repeatedly, and a 99-sensor dataset was obtained for every motion class. In the recognition step, random forest classifier was adopted. Finally, the recognition of the 12 motion classes of the tea-making procedure could be realized with greater than 0.98 accuracy, precision, recall, and F1-score.