![]() ThA1M1 In-person Regular Session, Room M1Ĭlassification of Pathological and Healthy Individuals for Computer-Aided Physical Rehabilitation This result suggests that WBED could immediately improve reactive balance function. The tlCOP of post-test significantly decreased in WBED group, compared to those in pre-test (p = 0.01). Total length of COP displacements (tlCOP) for 100 - 400 ms after the disturbance occurred were measured as pre and post intervention. Only WBED group subjects were disturbed in random direction by WBED during the tandem stance. All subject were asked to perform tandem stance exercise. The subjects were separated into WBED and sham group. ![]() Twelve healthy adult males participated in this study. Keywords: Rehabilitation Systems, Welfare systems, Human Factors and Human-in-the-LoopĪbstract: The purpose of this study is to investigate the effect of wearable balance exercise device for reactive balance function (WBED) on reactive balance function. Therefore, we applied feature elimination method based on cross-validation over players (leave one-player out) to extract features that are effective to classify the ball rotation and have high generalization performance for players.Įthologically Inspired Behavior Model for Sustainable Virtual Animal Assisted Activity The principal components can be roughly divided into those representing difference in players and those representing difference in rotation direction of balls. Since the frequency-domain features have overwhelmingly high dimension compared to the number of samples, dimensionality reduction was performed using principal component analysis (PCA). The acoustic features were extracted in the frequency domain, because it is expected to be more robust to noise than feature extraction in the time-domain. We adopt a ternary complete three-class support vector machine (SVM), which is considered to have higher generalization performance with small number of samples than deep learning. To deal with these problems, we employed a three-class classifier to classify rotation direction without ambiguity, and constructed a sound dataset with totaling 1506 balls from six players. In addition, since the sound dataset for training was constructed from a single player, the classification performance for different players was not ensured. ![]() Since the classifiers were trained independently, some balls could not be classified into the three ball types. In our previous study, we used three two-class classifiers to classify ball types with three different rotation directions: topspin, flat (low rotation), and slice (underspin). We have been developing a classification method for ball rotation direction from hitting sound, which provides a low-cost system without disturbing players. In tennis, it is an important skill for players to give a desired rotation to the ball when hitting and to detect the rotation direction when receiving. Keywords: Systems for Field Applications, Machine Learning, Human InterfaceĪbstract: In recent years, AI technology has been widely used to improve the performance of athletes, and this is no different in tennis.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |