Rehabilitation Action Evaluation Study Based On DTW Algorithm

With the continuous development of computer technology,human action evaluation has been actively studied in the last decade.Human action evaluation need calculate models to automatically detect abnormalities and evaluate the quality of human actions performed in specific postures,and related researches are widely applied in various fields such as rehabilitation medicine,assisted living,skills training and athletics.In rehabilitation medicine,as stroke,Parkinson’s and Alzheimer’s disease are often accompanied by motor dysfunction,early recovery of daily living ability is the main goal of patients’ rehabilitation,and specific rehabilitation action training is the main treatment means to improve limb recovery ability.In order to establish the link between rehabilitation training and action evaluation,this paper uses the Azure Kinect sensor to obtain human skeletal data,evaluate the daily upper limb action based on the improved dynamic time warping algorithm,and establish a scoring mechanism for action achievement.The main contents are as follows:(1)The Azure Kinect sensor is used to collect four kinds of daily action videos,such as drinking water,combing hair,touching the opposite shoulder and touching the back pocket,and to obtain the three-dimensional coordinate data of the key points of the human skeleton.Select the main key points,and eliminate the unimportant key points,and then fill vacancies and filter noise by using segmented cubic spline interpolation and mean filtering,respectively.(2)The three-axis variance sum and difference sum of the key points are calculated through the sliding window to find the peaks and valleys,and obtain the starting and ending points of the action segment for action segmentation.Establish the human structure model,construct three-axis space vectors of key points,calculate angles and distances between vectors,normalize features,and compose the action feature matrix.(3)The dynamic time warping barycenter averaging algorithm based on Euclid barycenter is used to make the only template for each kind of action.Finally,the traditional dynamic time warping algorithm based on feature matrix and the improved dynamic time warping algorithm based on feature weight are used to match the feature matrix sequence to be tested with the action template matrix sequence,and calculate the similarity between the two sequences.Referring to the Action Rating Scale,the scoring mechanism for action achievement is established to realize the action evaluation in healthy people and patients.The experimental results show that the similarity between the patient’s action data and the template’s action data is quite different,and the action evaluation method is feasible.Compared with the action evaluation of dynamic time warping algorithm based on feature matrix,the action similarity of drinking water,combing hair,touching opposite shoulder and touching back pocket in the action evaluation of dynamic time warping algorithm based on feature weight is improved,which can improve the performance of action evaluation.In the scoring mechanism for action achievement,the fitting result of the mathematical relationship between the action similarity and the action score shows that its confidence degree reaches99%,which can meet the needs of practical application.