In order to alleviate the great contradiction between the current stroke rehabilitation medical resource and the need of patient,reduce the pressure of medical staff,and realize the home rehabilitation training and evaluation of patient in the late stage of stroke rehabilitation,based on deep learning and machine vision technology,a post-stroke rehabilitation training system that can intelligently monitor the patient's completion of rehabilitation training action and judge whether the action is up to standard was proposed,so as to maximize the effectiveness of rehabilitation training and improve the recovery speed of stroke patient.The two-dimensional human joint point is recognized by deep learning neural network to improve the accuracy on the premise of ensuring the recognition rate.In the process of rehabilitation training,the patient's movement is monitored in real time by identifying the skeletal framework.The obtained joint point information is compared with the correct movement,the patient's incorrect posture is corrected,and the degree of completion is displayed.On the premise of ensuring the effectiveness of key point,this system has good real-time performance,fast response speed,the overall average confidence level reaches 98.2％,the number of video frames per second is between 15～30,and the accuracy rates of joint angle and completion number are more than 99％.Based on deep learning and machine vision technology,it provides a feasible solution for stroke patient to evaluate recovery and upper limb rehabilitation training without the guidance of medical staff.StageⅢ～Ⅴ stroke patients can complete rehabilitation training at home alone to alleviate the shortage of rehabilitation resource.
NETL
NSTL
万方数据
