Research On Gait Analysis System Based On Human Skeletal Point Data

Abnormal gait greatly affects the quality of daily life of patients.Therefore,analyzing gait,classifying and identifying abnormal gait can help improve the quality of life of patients and assist them in rehabilitation treatment.In thesis,a gait analysis system is designed and implemented by using the data of human body node provided by Kinect.The system mainly includes three functions: Kinect node optimization,gait parameter calculation and abnormal gait classification.Experimental results show that the system can effectively analyze gait and identify abnormal gait.The main work of thesis is summarized as follows:In order to solve the problems of low confidence and instability of the node in the case of Kinect self-occlusion,an optimization algorithm of the node was proposed based on encod-loop-decoder.In thesis,the time period to be processed is determined according to the confidence level,and the network structure of encod-circulation-decoder is adopted to process the time sequence data of the node with LSTM algorithm,so as to re-predict the location of the node with low confidence and complete the optimization of the human node data.Experimental Results The proposed method is suitable for walking movement,namely gait analysis,and is superior to the traditional interpolation method in dealing with different length time periods.In order to solve the problem of small range of segmented phase threshold in traditional threshold method in gait analysis and classification,automatic gait parameter calculation and abnormal gait classification scheme are designed.In the part of calculating gait parameters,thesis based on key point perception algorithm segmentation of motion state,automatic correction of walking direction,and on the basis of state segmentation combined with threshold to complete the division of gait phase,on the basis of phase division to calculate the temporal and spatial parameters of gait.In the part of abnormal gait recognition,according to the small amount of gait data in the system,machine learning method is adopted,and KPCA dimension reduction is adopted to integrate the spatio-temporal parameters and frequency domain characteristics of gait,and the abnormal gait recognition task is completed in the form of dichotomizing each gait.The experimental results show that the proposed method can automatically complete the calculation of gait parameters,and the recall rate of different gaits is above 85%,which can effectively complete the gait classification.Finally,a gait analysis system is designed based on the above method.The gait data of some elderly people were collected by the system as self-built data sets,which were divided into fall group and non-fall group.Through the experimental verification and functional test of the open and self-built data sets,the recall rate and accuracy rate of the recognition of normal gait were 87.72% and 84.75%,respectively.The experimental results show that the system’s functional integrity and algorithm feasibility can effectively complete the task of calculating gait parameters and recognizing gait,and provide gait analysis and early warning service for users.In addition,the system also provides account management,case display,results summary and other services for physicians and patients.