Research On Evaluation System Of Family Rehabilitation Training Based On Computer Vision

With the rapid development of society,population aging is rapidly becoming a global phenomenon,the incidence of chronic diseases is increasing,and an increasing number of elderly people are facing serious challenges with impaired physical functions such as muscle strength,balance and mobility.At present,the rehabilitation plan for the elderly in various countries’ health care systems is still in the initial stage.Routinal rehabilitation is carried out under the direct supervision of doctors in the hospital.After discharge,they return to the community or family for further rehabilitation exercise,so as to maintain and strengthen the effect of rehabilitation.However,in the family rehabilitation of many elderly people,the degree of adherence to the exercise program is low,leading to the extension of the treatment time,the rehabilitation effect is not good.Therefore,how to carry out effective exercise and rehabilitation exercise in the home environment has become an urgent problem to be solved and a research hotspot at present,among which the most difficult is how to make home rehabilitation equipment easy to use and how to accurately evaluate rehabilitation training.Aiming at the above difficulties in family rehabilitation,this paper proposes a family rehabilitation training evaluation method based on computer vision,and gives the system implementation.Ordinary cameras are used to collect the video stream information of patients’ rehabilitation training,so patients do not need to wear any equipment,thus greatly reducing the complexity of traditional rehabilitation training equipment.Posture estimation algorithm is used to capture the motion sequence of each node of patients,and data filtering,feature extraction and machine learning classification algorithm are used to score the movement of patients in rehabilitation training.A simple and easy to use rehabilitation training system with immediate evaluation feedback is developed,and the performance of the system is verified through clinical trials.This paper mainly studies the following aspects:1.Research and improve machine vision attitude estimation algorithmIn order to use ordinary camera to realize human pose estimation on home computer,a large number of improvements and optimizations were made on the basis of Carnegie Mellon University pose estimation algorithm OpenPose.The proposed algorithm achieves 39%accuracy after 39000 iterations in COCO2017 dataset,and achieves 30FPS real-time 2D attitude estimation on Intel(R)Core(TM)i5-9400 CPU@2.90GHz and NVIDIA GTX 1650 platform The e problem of real-time pose estimation of machine vision in family rehabilitation training is solved.2.Action scoring algorithm based on special normalization,dynamic time warping and support vector machineAccording to the characteristics of movement data in family rehabilitation training,a scoring algorithm based on special normalization,dynamic time warping and support vector machine(SVM)was proposed.The special normalization solves the position difference caused by the difference of human body shape or the difference of distance from the camera.The dynamic time warping algorithm solves the problem of the old people’s motion lag caused by the decline of motor ability.The evaluation model mapping with the evaluation of clinical experts is established by using the support vector machine algorithm.Through experiments,the accuracy of the motion evaluation algorithm was verified to be 93.3%,which showed a positive linear relationship with the scores of rehabilitation experts(r=0.967).3.Software development of family rehabilitation training evaluation systemOn the basis of the algorithm development,the family rehabilitation training evaluation system software is designed and developed with PyQt.The software functions mainly include login,registration,interactive training,movement evaluation and personal information management modules.Patients can be more actively involved in rehabilitation training,and at any time to view the individual training data and rehabilitation status.This research has proved the feasibility of using the algorithm based on computer vision to carry out rehabilitation training and motion evaluation for elderly patients with chronic diseases at home.The motion scoring algorithm in this paper can also provide reference for other related studies.