Research On Gait Analysis Method Of Rehabilitation Patients Based On Deep Learning

Gait analysis is a commonly used assessment method in the field of rehabilitation.In medical applications,gait data collection for patients uses marker-based systems.This system has high cost,high environmental requirements,time consuming process,and certain impact on patient's gait.Therefore,in the past few decades,more and more people have begun to study the marker-less human motion capture system.In addition,the current gait analysis results are displayed in the form of graphs.Interpretation requires certain professional knowledge,and the related classification tasks according to gait have become the research direction of many researchers.The purpose of this study was to develop a set of marker-less motion capture systems based on deep neural network(DNN)methods using multi-view patient's video.And based on the captured patient gait,the relevant causes are automatically classified.The system takes the video of the orthopedic patient as input.Firstly,the 2D human pose is estimated by the stacked hourglass network,and then the multi-view 2D body pose data is combined,and the 3D human pose is estimated by the DNN-based method.In the classification task,we divided the input gait video into three groups,namely,the left hip replacement patient,the right hip replacement patient,and the double hip replacement patient.The classifier takes the time series of the human body's 3D pose as input and automatically classifies the cause of the patient based on the long short-term memory network(LSTM)method.In the marker-less motion capture system designed for this study,the mean error for all patients was 29.87±4.99 mm.The average error of per patient is stable.In the classification task,the classification sensitivity of different categories is 58%~85% and accuracy is 70%.The marker-less motion capture system proposed in this study adopts multi-view information and has high precision.It can be used as a feasible way to replace the marker-based motion capture system in the future development.In this study,the classifier innovatively applied the long short-term memory network(LSTM),which has the most advanced results at present,and is a basis for applying deep learning for clinical gait analysis in the future.