Research On The Internet Of Things Action Monitoring System For Smart Fitness

In recent years,with the increasing intensity of work and pressure of life,people's health is facing many challenges.In this context,healthy living has become a topic of concern.On the other hand,the rapid development of electronic technology encourages people to use MEMS-based inertial sensors to monitor individual sports,thereby improving the effectiveness of physical exercise and promoting physical health.At present,researches on MEMS-based motion recognition have the following characteristics:in terms of data transmission,Bluetooth and Zigbee are widely used.However these methods are applicable to personal area networks and do not support multi-user scenarios.In terms of motion recognition algorithms,classical machine learning algorithms are mainly used,such as support vector machines,naive Bayes,decision trees,etc.However,Deep neural networks that have strong understanding ability have not been used.In terms of application scenarios,researches are mainly applied to human activity recognition,that is,to recognize the state of standing,lying,sitting,riding,etc.There are few studies on recognition for specific limb motion.Aiming at solving the above problems,this thesis designs and implements a MEMS-based multi-user motion monitoring system.The system is used to obtain the three important parameters of motion,that are,the type of motion,the number of motions,and the cycle of motion.The specific research contents of this paper are as follows.(1)In terms of systerm design,considering the scenario for multi-users and the accuracy requirements of motion perception,the ESP8266-based Wi-Fi transmission system and the MPU9250-based attitude acquisition system have been deeply studied.After considering factors such as power consumption and serial communication,a reliable wearable hardware solution is designed and implemented.(2)In terms of algorithms,study on motion pattern recognition and frequency analysis algorithms was performed.For action pattern recognition,two kinds of classification algorithms with different complexity are proposed.These two algorithms are respectively based on SVM and deep neural network(1D CNN and LSTM),which can be adapted to scenarios with different computing power.For frequency analysis,a method based on zero-crossing detection and wavelet transform is proposed to count the number of motions and calculate the period of each motion.Through the motion recognition algorithm and the frequency analysis algorithm,the calculation of the motion type,the number of motions and the motion cycle is realized,and a comprehensive description of the limb motion is obtained.(3)In terms of system development and experimental verification,a hardware platform based on the action collection bracelet and a software platform integrating data receiving,motion recognition and cycle calculation are developed.The software platform can be deployed in the cloud server.The data acquisition bracelet,pattern recognition and cycle calculation algorithm and system software form an experimental platform.The experimental results based on the software and hardware platform show that the data obtained by the attitude acquisition system can be correctly transmitted to the host computer through Wi-Fi.Experiments based on 7 types of-upper and lower extremity movements show that the proposed deep neural network has a good learning effect on small data sets,achieving 97.61%accuracy of motion recognition,and SVM has reached more than 96%recognition accuracy.In the 50-time action frequency experiment,the frequency statistics algorithm achieved 100%calculation accuracy,and the motion cycle calculation result was also close to the real value,which proved the validity of the cycle calculation method.The system provides a multi-user communication method,which achieves accurate and reliable human motion monitoring,and has broad application prospects in the fields of physical education and rehabilitation training.