Gait Analysis Based On Smart Insole And Its Applications

As the aging of the population continues to increase,the incidence of some diseases has increased significantly in recent years,so early prediction and auxiliary diagnosis of these diseases is particularly important.The symptoms of many diseases are manifested in the gait of continuous walking,which affects the pace of walking,stride,stride frequency and plantar pressure.With the development of the wearable health-monitoring systems(WHMS)in recent years,it's simple to collect the gait information while walking,and then it can achieve the purpose of health monitoring and auxiliary diagnosis.Based on wearable smart shoes,this study collected gait data-sets containing four abnormal gaits and one normal gait.Two gait recognition methods based on deep learning are studied,recognition of different gaits using Long Short-Term Memory(LSTM)and 1DConvolutional Neural Network(1D-CNN).Research contents are as follows:The etiology and characteristics of various abnormal gait were studied.Four abnormal gaits were selected according to the principle that the disease has commonness,predictability and regularity.The various steps required to preprocess gait data are described.This paper introduces how to extract a single gait segment from the continuous walking process,extracts a single five-time feature node from a single gait according to the walking characteristics of normal people,and divides the continuous gait accordingly.Specific analysis of the four abnormal gait and a normal walking gait to ensure the accuracy of the segmentation results.Two deep learning algorithms for processing time series were studied.The structure and training process of LSTM are introduced,and what improvements LSTM has over the original RNN to accommodate longer time series are introduced.It introduces the structure and training methods of CNN and how to improve CNN so that it can process one-dimensional time series data.According to the normal standing and walking acceleration and the plantar pressure,a reasonable sensor layout was selected,and a dataset containing five gaits was collected.The LSTM and 1D-CNN algorithms were built using the PyTorch deep learning library to identify five gaits.The results show that both algorithms can achieve a good recognition effect.