Research On Human Posture Recognition Based On Wearable Sensor

With the development of science and technology,human posture recognition technology has made great progress.It is widely used in medical engineering,entertainment games,film and television production and other fields.In order to study human body posture,a set of wearable equipment is made by using acceleration inertial sensor,which is used in the research field of human body attitude recognition.Attitude recognition based on wearable inertial sensors is different from recognition based on video and optical sensors.It has better recognition rate and practicability.This research system is mainly divided into two parts,the first part is the human posture data acquisition subsystem;the second part is the upper computer human posture recognition subsystem,including data processing,recognition and database.In this system,the data acquisition subsystem mainly installs the nine-point inertial sensors in nine parts of human body to collect the human posture data,and then sends them to the receiver.The human posture recognition subsystem of the upper computer is through receiving the data of the data acquisition subsystem,processing the data,and identifying and classifying the corresponding posture.First of all,this paper analyzes the needs of the data acquisition subsystem,and determines the main framework of the subsystem.Then a simple analysis of human kinematics is made to determine this system for the acquisition data of accelerometer.Through a simple analysis of the various human movements,combined with the human rigid body model,the data acquisition format package is established;and through experimental analysis to determine the use of nine points,that is,the nine parts of the human body for data acquisition.A pseudo-RGB data fusion method is proposed.Through collecting the acceleration sensor data of 9 parts and sending the data by wireless serial port in the form of data packet,the human body attitude data acquisition subsystem is composed.Secondly,this paper analyzes the requirements of human posture recognition subsystem,and realizes five parts: data receiving,data processing,network training,action recognition and building database.The visual program interface is established by using Python language.The data is received by the wireless serial module of PC and saved as a CSV data file,and data processing is done by noise reduction and extracting time-frequency domain information.The improved convolution network algorithm is implemented by using the TensorFlow deep learning framework under Python to save the trained network model.The saved network model is analyzed and the target data is input to identify the action.Then,the improved convolutional network based on residual network is proposed in this paper.Compared with the traditional convolutional network,the improved convolutional network is more effective,and its recognition rate can reach 98%.Finally,the whole system is tested,and the system recognizes various positions such as sitting,running and up the stairs,and achieves the expected goal.