Design And Research Of Human Standing Posture Recognition Algorithm Based On Flexible Sensing

With the progress of society,the new trend of life style has brought great changes to people’s life.Poor posture and its associated symptoms are becoming an important public health problem.Abnormal posture is a kind of bad posture.At present,there are mainly two forms of human posture recognition based on image and sensor.However,the human posture recognition based on image is mostly limited by environmental factors.In order to solve these problems,this paper intends to convert the numerical value of foot pressure directly into the input data of deep learning network by means of flexible sensor array,and combine the deep learning algorithm to recognize human standing posture.This paper focuses on the research of sitting posture recognition technology based on flexible sensor,and carries out four aspects of work:(1)Three kinds of pressure sensor structures with different electrode arrangement were designed and fabricated,and the performance characterization test was carried out by using relevant instruments.The SS-3 electrode arrangement was selected as the sensor device structure used in this study.Then through the performance test of it,the problems will be simulated in the use process.Based on this Sensor unit,an All-Textile Pressure Sensor Array(ATPSA)with 1024 sensor points is designed.(2)In order to better collect human stance feature information,ATPSA needs to arrange as many sensing points as possible.Therefore,it is necessary to formulate a high-speed signal acquisition scheme in the scene of the whole fabric based flexible sensor array,which provides a data analysis basis for later data imaging and stance recognition.(3)The data visualization software was written for the completed ATPSA,and the imaging effect was tested.The stance is divided into six common categories,and the stance data of different stances are collected by volunteers and the corresponding stance data set is established.Then,a three-layer BP neural network was constructed to verify the feasibility of the ATPSA acquisition data for extracting foot feature information.(4)In order to improve the accuracy of stance judgment,the existing convolutional neural network model Le Net is used,and the relevant parameters are fine-tuned to build a Le Net model more suitable for the data set format of this paper.The generalization ability of four different models(SVM,RF,BP,Le Net)is experimentally compared.Finally,it is concluded that ATPSALe Net,the general system for human posture monitoring and evaluation proposed in this paper,is most suitable for extracting stance feature information.This paper proposes ATPSA-LeNet,a new universal system for human posture monitoring and evaluation with high robustness.Using this classifier and sensor fusion technology,foot pressure values can be directly converted into input data through ATPSA,and the nonlinear correlation between plantar pressure distribution characteristics and foot morphology can be established.The accuracy of stance recognition is 98.84%.The experimental results show that ATPSA-Le Net can provide reference for clinical diagnosis and rehabilitation process,which has great significance and application potential.