Research On Table Tennis Movement Pattern Recognition Based On Acceleration Sensor

In view of the relatively few research work on pattern recognition of table tennis currently,and problems such as bad real-time performance and low recognition accuracy,this thesis use ESP32 microprocessor integrated with MPU-6886 acceleration sensor as the acceleration signal collection and transmission device,and studies the pattern recognition for four common actions in table tennis.The sports of table tennis is complex,how to realize the real-time collection and transmission of acceleration signals during table tennis,design effective feature engineering,select an appropriate recognition classifier and improve the generalization ability of classifiers on different sports individuals are all issues to be considered.Therefore,this thesis conducts the following research work:1.Acceleration sensor is used to collect acceleration data generated by users during table tennis,and at the same time,the data is transmitted to PC for processing and classification.Traditional machine learning algorithms(KNN,Support Vector Machine,Decision Tree)are applied to table tennis swing action recognition,and a real-time classification algorithm based on decision tree is designed.The main processes include: Firstly,using the sliding mean filter algorithm to filter the data noise of the original acceleration,and implement windowed segmentation of filtered data.Secondly,22 dimensional time domain features are extracted,and then the features were dimensionally reduced by Principal Component Analysis.17 main features are selected and generated based on the cumulative contribution ratio.Finally,the data processed by feature engineering is used as a sample set,and decision tree algorithm is used to build the motion classification model by pruning and adjusting parameters.The results show that the decision tree classification model designed in this thesis can effectively realize the real-time recognition of the four sports modes of table tennis,and the real-time recognition accuracy can reach 97.32%.2.A table tennis pattern recognition method based on Long Short-Term Memory networks(LSTM)model is designed.The pre-processed acceleration data is directly input into the double-layer LSTM network model,which can automatically extract the time series feature of the input data and recognize the motion pattern.The results show that the accuracy of this method can reach 97.10% in identifying the four sports modes of table tennis.In addition,this thesis also compares the differences between LSTM and traditional machine learning algorithms in the process and performance of table tennis action classification,and verifies the effectiveness of this method by comparing with the relevant research results in recent years...