Research On Classification And Level Assessment System Of Tennis Basic Technique Based On IMU And Machine Learning

Background:Tennis is heavy depended on technical movement.In daily tennis training,the effective monitoring of technical movement quality is of great significance to arrange the training load of tennis players scientifically.The current tennis technical movement quality evaluation is generally determined by the coaches subjectively,and there is a problem that the training effect of multiple players cannot be feed-back in real time.Purpose:This research used the inertial measurement unit(IMU)collected tennis strokes data,process the data by machine learning methods,thus real-time recognition and classification of five basic tennis technical movement at three different levels subjects.One way ANOVA results based on the three levels of different level subjects ITN test,technical movement visual image contrast and classification results to establish evaluation system.Therefore,it can provide real-time biomechanical data feedback for coaches or players.Subjects and methods:subjects:36 male right-handed domain subjects were recruited and divided into elite(N=12),sub-elite(N=12)and amateur group(N=12)according to their sports performance.Methods:1.ITN:ITN test was strictly conducted in accordance with the corresponding standards,and one-way ANOVA was conducted on the training experience of the subjects at three different levels and the results of ITN field test.2.Data collection:The IMU(sampling frequency:50 Hz)worn on subjects with domain hand wrist,including tri-axis acceleration and tri-axis gyroscope sensors,and it through wireless low-power Bluetooth technology connected smart phones to collected participants’ tennis serve,forehand,forehand volley,backhand and backhand volley each 30 effectively strokes data.Then the data will be transmitted to the terminal in real time through cloud storage technology for data processing.3.Data processing:IBM SPSS 26.0 was used for one-way ANOVA of ITN test results.After collecting the inertial measurement data,the tennis technical movements were recognized and classified by data preprocessing,segmentation,feature extraction,dimension reduction and classification using the Support Vector Machine(SVM),K-NN and Naive Bayes(NB)machine learning algorithms.Results:There were significant differences in training experience,ITN field test scores,grades and specific tests among different levels participant(P<0.001).The raw data of tri-axial acceleration and tri-axial angular velocity can clearly distinguish the five types tennis technical movement of the subjects at three different levels.The average accuracy of the SVM(StandardScaler)in recognize and classifying the tennis technical movement of the subjects at three levels was up to 90%.The accuracy of the K-nearest neighbor algorithm is the highest when K=7,but only reaches 47%.The Naive Bayes algorithm achieves the highest accuracy of only 64%.Conclusion:This study confirmed that IMU worn at the wrist of the subjects’ domain hands and connected with smart phones can effectively collect the tennis stroking data,and the SVM(StandardScaler)can recognition and classify the five tennis technical movements of the three levels subjects with an accuracy of 90%.Moreover,tennis level evaluation system can be built based on IMU equipment and machine learning.