Research On Soccer Motion Recognition Algorithm Based On Wearable Devices Of Internet Of Things

With the development of Io T technology,machine learning and other artificial intelligence technologies,there have been many technologies applied to the sports industry.As the first sports in the world,football or soccer is widely popular,highly regarded and commercialized.In this context,there is a great demand and potential for combining Internet of Things(Io T)and Artificial Intelligence(AI)technologies training and competition of soccer.Soccer motion recognition and analysis mainly focus on motion capture and evaluation of the lower legs and ankles of athletes.Wearable devices with built-in inertial sensors are strapped to the ankles,and machine learning algorithms are used to recognize motion movements and estimate motion intensity.In general,the sensitivity of inertial sensors for data capture of lower body motion movements is low,especially in soccer where the movements are complex and individual athletes vary greatly,and existing algorithmic models for human activity recognition research are difficult to be used directly for soccer movement recognition.The traditional training methods of soccer sports have limited collection and mining of real raw data using wearable sensory devices,and lack of human movement capture and gesture recognition based on sports science theories.Current models do not analyze complex and high noise action data streams deeply enough.To address the above shortcomings,this thesis proposes a framework for soccer motion recognition and evaluation based on artificial intelligence and Io T.It uses wearable devices to collect soccer motion data and constructs machine learning algorithm models to recognize and evaluate soccer motion as well as analyze motion intensity.The main contributions and innovations of this paper are as follows.1.Common feature extraction methods are suitable for recognition of most monotonous movements,but soccer movements are highly complex and the athletes’ ankle movements are flexible and changeable.The actual acquired soccer data streams are noisy and the data patterns are not obvious,and the performance of the model for data recognition will be degraded.In order to extract the effective feature values in the complex data stream and improve the model recognition accuracy,we propose the angle solver with SVM classification model,which is suitable for the recognition of soccer moves because the small sample data is less affected by noise.2.The proposed angle solver with SVM classification model can accomplish soccer motion recognition and quality assessment in high accuracy,but the characteristics of the SVM algorithm make the proposed classification model unable to handle the data stream directly,and a single algorithm model can only accomplish a single classification task.To reduce the delay of the algorithm output results and the resource consumption caused by data operations,a multi-task learning model based on the LSTM algorithm is proposed to achieve action recognition and motion intensity estimation.The model can perform classification subtask and regression subtask in parallel and output the results simultaneously.In terms of initial data processing,a feature extraction scheme is designed based on the original IMU sensor data,and feature data augmentation collection is performed to solve the small sample data problem.3.The existing sensor-based public soccer sports datasets are mostly group sports datasets.To evaluate the performance of the proposed soccer action recognition algorithm,this thesis designs a data extraction experimental scheme to complete the data collection of different actions and analyzes the original data characteristics for data cleaning.The experimental results of the stance angle solving SVM classification algorithm model show that the accuracy of the proposed action recognition algorithm model based on sensing technology can reach 90%in recognizing different motion;the action quality assessment model can classify professional athletes and amateurs with an accuracy of 93%.In terms of validating the multi-task model based on deep learning algorithm,in order to explore the generalization of the model,three publicly available datasets were used for model validation,and the characteristics and learning ideas of the model were analyzed.Then experiments are conducted on the collected soccer sports action datasets,and the experimental results show that the proposed multitask model can perform two subtasks simultaneously and excellently.