Basic investigation of a method of assessing tennis forehand stroke quality using convolutional neural networks and inertia sensors

To improve the movement performance of athletes, it is desirable to have an environment in which the movement quality ca be accurately assessed. For that reason, in this study, we construct a model to assess quality using inertia sensors (acceleration/angular velocity signal measuring devices) and convolutional neural networks (CNN). The forehand stroke in tennis was selected as the subject movement for quality assessment. We took data from one player who had previous experience in playing tennis and two who had not, and labeled data from the experienced player (30 samples) as high quality, and labeled those from the inexperienced players (30 samples) as low quality. We divided the total of 60 samples into 40 samples of training data and 20 samples of test data, constructed the CNN based on the training data, and tested the estimation accuracy using the test data. The result displayed 95.0% of correct answer rate, demonstrating high estimation accuracy. However, due to the low number of test subjects, it shall be necessary to increase this moving forward, to get more reliable results.