Study And Design Of A Wristband Low Power And High-Precision Ping-pong Recognition And Analysis System

With the development of artificial intelligence and Internet of Things to promote the intelligentization of wearable products,smart wearable devices are continuously combined with sports and health to find new breakthroughs.Almost all smart bracelets on the market are universal bracelets.These products provide a single function that does not provide analysis and guidance for sports enthusiasts.Therefore,the development of a smart bracelet with motion recognition will have a better market application prospects.In this thesis,based on the study of the characteristics of ping-pong,a low-power,highprecision ping-pong system was developed by adopting the combination of emerging artificial intelligence technology and traditional hardware technology,which includes a real-time recognition analysis ping-pong bracelet and handheld terminal APP.The bracelet is responsible for real-time recognition of the motion state and sends the recognition result to the APP through Bluetooth,which is composed of a sensor module,an MCU main control module,a Bluetooth module and a bracelet mechanical structure.The APP is responsible for statistics and analysis of exercise status which is composed of a motion state statistical software module and an analysis software module.In order to achieve the high precision of the system,aimed at the three stages of ping-pong including action preparation,hitting and action end,an action endpoint detection algorithm was proposed in this thesis based on the abrupt change of the action signal.The action is divided into a motion start transition segment,an action segment,and an action end transition segment,the starting point of the motion is extracted according to the relationship between the difference of the before and after motion signal value,the fixed threshold and the current state.Since the human body's kinetic energy is concentrated at 0~10 HZ,in order to obtain effective features,the action signal was decomposed into three layers based on db4 wavelet,and six kinds of actions was classified by the genetic algorithm-optimized S_Kohonen(Supervised Kohonen Networks)neural network,which include forehand push,backhand push,forehand draw,backhand draw,forehand croquet and backhand croquet.compared with the single S_Kohonen neural network,the results show that the method can improve the recognition accuracy of the system.In order to achieve low power,the hardware circuit was designed by low-power devices including STM32L432 KC,MPU6050 and DA14580.In the algorithm implementation,a series of algorithm were optimized by the Floating point unit(FPU)and DSP instruction library of STM32L432 KC.Compared with the traditional method,the system power consumption is 80 mW,the recognition time is 21 ms,after optimizing the system power consumption is 14 mW,and the recognition time is 2ms.The results show that the way can significantly reduce the power consumption of the system.The size of the bracelet developed in this thesis is 42.5mm×36.5mm.With a rechargeable design,the standby and operating mode power consumption of the system is 0.28 mW and 14 mW.The experimental results show that the offline average recognition rate is 99.1%,and the real-time average recognition rate is 92%.