Action Recognition And Its Application In Upper Limb Rehabilitation Training

As the degree of aging at home and abroad to deepen gradually,the health problems of the elderly have received widespread attention.Stroke is a disease with the highest incidence and greater harm,and the lack of upper limb motor function caused by its sequelae has become a hot research topic at home and abroad.At present,in response to the problem of patients with unilateral upper limb motor function loss,one method is to assist the patient in rehabilitation training through a rehabilitator,but the labor intensity is too high;another method is to use rehabilitation robots for assisted training.However,traditional rehabilitation robots can only drive patients to perform single and repetitive exercise training.The patient’s sense of participation is low and human-computer interaction is poor.In order to solve these problems,the movement of the patient’s healthy limb is recognized based on the fusion of acceleration and surface EMG signals(sEMG),and the movement trajectory is planned according to the identified movement.The upper limb rehabilitation robot is controlled to drive the affected limb to track trajectory,it is to realize the function of the patient’s own healthy limb to drive the affected limb for training,and then realize the role of liberating rehabilitation practitioners,enhancing patients’ sense of participation,and improving human-computer interaction capabilities.The main work of this subject is as follows:(1)The acceleration signal and sEMG signal are collected and preprocessed.For the experimental platform,5 males,5 females,10 subjects and 7 actions are selected for simultaneous acquisition of acceleration signals on two axes and sEMG signals on six channels.Kalman filter method is used to denoise the acceleration signal.A wavelet denoising method with improved threshold function is proposed to remove the baseline drift and high frequency noise of sEMG,and solve the problems that cannot be solved by traditional wavelet denoising.The frame method is used to extract the effective signal segment to reduce the amount of processed data.(2)Actions are recognized based on the fusion of acceleration and sEMG signals.Two features in time domain and frequency domain are extracted from acceleration signal and 3features in time domain and frequency domain are extracted from sEMG signal.The DBI evaluation index is used to evaluate the similarity of features to solve the situation of large number of features and high dimension.The best fusion feature PSD+SK+MAV+MPF is selected for classification and recognition.The C and g parameters of SVM are optimized by particle swarm algorithm.The recognition rate and running time are used as indicators to verify the effect of the classifier,and it is proved that PSO-SVM is better than SVM and GA-SVM.After the classifier classification,the effective recognition of 7 kinds of actions has been realized,and the action recognition rate has reached 92.86%(3)Research on the control is developed about upper limb rehabilitation robot based on action recognition.By establishing a two-link model of the affected limb,the trajectory of the affected limb is planned according to the relationship between the joint angle and the end position.The motor speed is controlled,and the end trajectory is tracked by the rehabilitation robot.Fuzzy PID algorithm is introduced to realize the rapidity and accuracy of trajectory tracking.And is carried out to the tracking simulation of the motor speed and the end position of the planned trajectory.Control methods and strategies are proven feasible based on experimental verification.(4)Control research based on action recognition is carried out on the upper limb rehabilitation robot platform.The Fourier M2 upper limb rehabilitation platform is used to design the motion recognition experiment of the healthy limb and the experiment of the rehabilitation robot driving the affected limb to track the trajectory.And the online recognition rate of the 7 action is 91.6%,using the PSO-SVM model,which proves the high accuracy of the two signal fusion for action recognition.Fuzzy PID is used to achieve accurate tracking of the planned trajectory,the goal is achieve for using a healthy limb to drive the affected limb for rehabilitation training,the rehabilitation practitioner are liberate,and the patient’s sense of participation and human-computer interaction are enhance.