Research On Calibration Method And Human Motion Recognition Based On MEMS Accelerometer In FES System

The incidence of stroke in China is one of the highest in the world.Due to the country's high attention,the prevention and treatment of cerebrovascular disease in China has achieved initial results,but stroke is still the leading cause of death and disability among adults in China Therefore,the prevention and treatment of stroke in China is urgent and the patient's rehabilitation system needs to be improved.In the process of recovering the limb function of stroke patients,it is necessary to wear rehabilitation equipment equipped with MEMS sensors to assist in rehabilitation training for some specific movements.This article focuses on improving the accuracy of MEMS accelerometer and using MEMS sensor data to identify human movements and postures.A series of research work(1)In order to ensure the accuracy of the acceleration sensor data and not affect the accuracy of motion recognition,we need to calibrate the acceleration data.Under this background,this paper starts from the error analysis,establishes a parameter calibration model,proposes a method to calculate the automatic calibration gain by using redundant acceleration vectors,and then uses the calculated gain to perform the data on the acceleration sensor calibration.Finally,the accuracy and robustness of the proposed calibration method are verified by experiments.(2)At present,the research of upper limb pose recognition is still in its infancy.Due to the diversity of objective environment and the complexity of human pose,there is no public data set for upper limb pose.In this paper,an upper limb data acquisition system is designed.The MPU6050 sensor module is used as a data acquisition device.The three-channel data acquisition method is used to collect acceleration signals and gyroscope signals as sample data The collected acceleration and gyroscope data are sent to the host through a serial port connection.After the host receives the data sent by the three sensors,it sends the data to the computer through the serial port.After the collection is completed,the data set is preprocessed such as deduplication,interpolation and feature extraction,which integrates more comprehensive limb posture information and provides data guarantee for subsequent posture recognition work.(3)Experiments on KNN,logistic regression,and stochastic gradient descent algorithms are performed with the goal of recognizing human motion.In order to verify the superiority of each algorithm,the data window was adjusted and the recognition speed,calculation time and accuracy of each classifier were compared.Aiming at the problem of improving the accuracy of human pose recognition,a fully connected neural network model was established.In the process of constructing the network model,this paper studied different activation functions and optimizers.After experimental comparison and analysis,the softplus activation function and adagrad optimizer with better recognition performance were selected.Finally,by comparing with other classification models,comprehensive recognition accuracy and time efficiency are verified,and the superiority of fully connected neural networks in human motion recognition is verified.