ABody Motion Comparison Method by Combining Posture Evaluation Operators

In this paper, we propose a body motion comparison method by combining posture evaluation operators to compare appropriate posture between a trainer and a trainee in the various sports. The features of this proposed method are that to set basic evaluation formulas for time-series data as posture evaluation operators, and to evaluate human posture by combining the operators. The originality of this method is that instead of defining items to evaluate characteristics of poses in each sport, only the basic elements of body posture, such as angle, balance, and timing, are used for evaluation. In this method, posture feature data that is obtained from videos is used as input data. The input data contains the coordinates of 14 points of human body joints. The defined operators calculate each evaluation value on the human body joints at each time. In this method, the evaluation operator is selected for each sport, called context. A constant interval time for evaluation of time-series data, called the window size in this method, is changed by the result of Discrete Fourier Transform. The posture feature data is obtained from video of human body motion by using MediaPipe. This paper describes the details of this method and its current status to show the effectiveness and feasibility of it on bending and stretching exercises. We assume that a personal training system as an application of this method in order to support those who want to maintain their good healthy.