Research On Table Tennis Motion Recognition Based On Smart Watch

With the rapid development of information technology,the computers used by humans are more diverse,miniaturized and systematic.People and computers are interdependent in harmony,and computers are integrating into every aspect of people's lives.The emergence of a variety of new smart wearable devices such as smart bracelets,smart watches,smart glasses,and smart clothing has brought many changes to people's lives.Wearable computing is gradually emerging as a hot research direction.A wide variety of wearable devices are playing an increasingly important role in sports monitoring,medical health,and social entertainment.Wearable computing covers a wide range of fields and contains many kinds of technologies,and the study of human-computer interaction(HCI)is an important topic,which plays a fundamental role in many aspects.There are many ways of HCI.From simple interactions such as input and output to today's interactive forms based on voice,image,wireless signals and various sensory information,human-computer interaction technology is generating new vitality in the advancement of wearable computing.The recognition of human motion is a key technology in the research of wearable computing and human-computer interaction.It plays a vital and fundamental role in understanding human behavior intentions and harmonious interaction with the environment.For example,in the development of virtual reality games or auxiliary training systems in sports,it is especially important to accurately recognize the motions.At present,the methods of action recognition include computer vision based,inertial sensor based and wireless signal based.The development of computer vision technology is relatively mature,but the equipment needs higher precision.It is greatly affected by light,and its range of perception is limited,it needs to work within a certain distance.Common inertial sensors include accelerometers and gyroscopes,which have been studied by many scholars.It can overcome the shortcomings of visual technology,but it needs to wear the inertial node on the designated part of the human body.It is very inconvenient to carry it,and it will bring a lot of extra burden.Wireless signal-based sensing is now developing,its effects are still not ideal,and the devices is expensive,which is not conducive to widespread deployment.A variety of environmental resources,a pervasive and convenient motion recognition technology can accomplish a specific task more harmoniously.A wide variety of wearable devices can integrate a variety of sensors including inertial sensors.The motion recognition research of general wearable devices usually requires the user to carry multiple node devices,which brings a lot of extra burden to people.Some wearable devices or wearable fabrics are difficult to integrate multiple sensors that can be effectively used at the same time,which makes it difficult to study motion recognition in many practical situations.Especially in sports,the accurate recognition of movements with a pervasive and convenient daily device can not only effectively promote the development of relevant theoretical techniques,but also have important significance for the study of many practical situations.As a wearable product that is becoming more and more popular,the smart watch integrates various sensors inside,which plays an increasingly important role in activity-related measurement and physiological monitoring.By effectively utilizing its sensing information,it is possible to recognize human motions in a specific scene.The motion recognition based on the smart watch can effectively overcome the shortcomings of the visual technology,and at the same time give full play to the advantages of the inertial sensor,so as to complement each other.It has the unique advantages of convenience,reliability,stability and small environmental dependence.In addition,its ability to communicate in a variety of networks enables it to be better integrated into the information environment.This not only can fully exploit the advantages of the Internet of Things technology,but also effectively overcome the current lack of computing resources and small memory of the wearable device to achieve a good deployment and user experience.Table tennis is a skillful game with fast movement and high flexibility.Different from the recognition of other human behaviors,the recognition of table tennis movements has its inherent characteristics and difficulties.In this paper,a common commercial smart watch is used to study the recognition of table tennis movements.Accelerometers,gyroscopes,and magnetometers integrated in the watch are used.In order to obtain the data,this paper designs and develops a complete data acquisition system based on the watch-mobile phone-server architecture,which can stably obtain a total of nine-axis sensing data including acceleration,angular velocity and magnetic induction.In order to detect the human motion in the continuously transmitted signal,this paper proposes an motion signal segment extraction algorithm(MSSE)based on the signal variance,which can effectively detect and extract the motion signal segment.Then using the time domain analysis and frequency domain analysis of the signal,and the related theory of machine learning,a motion recognition method based on signal analysis and ensemble learning(SAEL)is proposed.In order to effectively utilize the local information of the signal and improve the generalization performance,this paper proposes a motion recognition method based on convolutional neural network signal feature optimization(CSFOR).In order to effectively utilize the local information of the signal and improve the generalization performance of recognizing more complex motions,this paper proposes a motion recognition method based on convolutional neural network signal feature optimization(CSFOR).In the experiment of eight basic table tennis skills,the accuracy of SAEL and CSFOR reached 99.41% and 98.24%.In addition,we have done a series of experiments to explore the performance differences of each sensor in table tennis motion recognition.The results show that the magnetometer can play a greater role.