| With the rapid development of science and technology now,the social rhythm is accelerating,the elderly population is increasing,the incidence rate of stroke,lumbar disc herniation and other diseases is increasing year by year,and the number of people with limb movement disorders caused by accidents is increasing,which seriously affects people’s limb functions.How to enable patients to enjoy the dividends of technological development and meet their daily home rehabilitation training needs has become one of the current research hotspots.Traditional rehabilitation training is conducted in hospitals,community health service centers,and other places,requiring the assistance of rehabilitation specialists.However,rehabilitation training systems based on rehabilitation training robots and sensors require high equipment requirements and are difficult to operate.Therefore,based on computer vision,this article proposes a method for evaluating rehabilitation training actions using ordinary cameras,without the need to wear equipment or purchase 3D cameras,greatly reducing the complexity of rehabilitation training.The main work of this article is as follows:(1)Sort out the background and current status of rehabilitation training research,review relevant theories and methods,shoot patient rehabilitation training actions and standard template action videos,and use pose estimation algorithms to extract key bone point information,obtain time series of action bone points,and construct the dataset for this study.(2)A 2D image motion matching method with human orientation feature is proposed.To compensate for the lack of Z-axis coordinates in 2D images and their impact on motion matching,a detection method for orientation between the human body and the camera is proposed;A new similarity measurement method based on limb partitioning has been proposed,which adds time constraints and improves the DTW algorithm to solve the problem of matching failure caused by different patient unrelated limb regions and standard template actions,as well as the problem of large temporal differences in matching actions.Through comparative experiments,analyze the impact of human orientation features on the matching accuracy of DTW algorithm;The impact of common similarity measurement methods and the similarity measurement method proposed in this article on DTW performance;The impact of increasing time constraints on the matching accuracy of DTW algorithm;Comparison of the performance of improved DTW algorithm compared to other matching algorithms.The experimental results show that the improved method in this thesis improves the accuracy of matching by about 9.7 percentage points compared to using the original DTW algorithm.(3)Propose a method for evaluating rehabilitation training actions.In order to solve the problems of difficulty in evaluating the training actions of patients in the early stage of rehabilitation in evaluation methods,lack of specific evaluation results for individual actions,correlation with expert experience,and medical scientificity,a new method and calculation formula for evaluating rehabilitation training actions were designed based on action matching.The experimental results show that the Pearson correlation coefficient r between the system score and expert score is 0.925,ranging from 0.8 to 1.0,with a strong positive correlation,indicating that the evaluation method can provide feedback on evaluation results consistent with expert experience.(4)On the PyCharm platform,use PyQt5,Spring Boot,MySql,etc.to build a rehabilitation training system,design system functions,and guide and evaluate patients’ training movements.The experimental results show that the average accuracy of the system rating is 93.77%,and the Pearson correlation coefficient r between the system rating and expert rating is 0.925,which is between 0.8 and 1.0.It has a strong positive correlation,indicating that the evaluation method can provide feedback on evaluation results consistent with expert experience. |
PDF Full Text Request