Research On Key Technologies Of Virtual Rehabilitation System Of Hand Based On Dynamic Gesture Recognition

The human hand is known as the second brain of human beings.In normal human life,most of the fine movements are done by the hands.However,in recent years,with the change of people’s diet structure and the acceleration of work rhythm,cardiovascular diseases have become the first disease that endangers people’s health,and this disease is also the main cause of hand dysfunction.However,the existing hand virtual rehabilitation system is boring and lacks guidance and fun,and even brings secondary injuries to patients in the process of rehabilitation training.In view of these problems,there is an urgent need to design a more effective,safe,and interesting hand virtual rehabilitation system.The main research contents of this paper are as follows.1.To address the problems of poor accuracy of dynamic gesture recognition and long training time of the model in the hand virtual rehabilitation system,a three-dimensional convolutional network model combining spatial attention mechanism and residual network is proposed.First,in order to obtain feature information of hand images,the network depth is increased by using Res Net residual network,which can ensure the model depth to obtain feature information and improve the classification accuracy of the model at the same time.Second,the spatial attention mechanism module is introduced to make the proposed network in this paper focus more on the target regions that need to be focused.Experiments show that the proposed model achieves 93.13% classification accuracy and achieves a better performance performance.Finally,the width learning model is introduced into the existing network to reduce the training time of the network.The comparison experiments with other classification models show that the training time of the proposed model in this paper is significantly lower than other classification models,which ensures the real-time performance of dynamic gesture classification as well as a high classification accuracy.2.To address the parameter optimization problem in the SVM model,a hand muscle motility classification model based on the improved SVM model parameter optimization is proposed.Firstly,Bio-capture electromyographic sensor is used to collect hand EMG signals;secondly,the generation mechanism of EMG signals,signal denoising,feature extraction method and feature dimensionality reduction by principal component analysis are analyzed;finally,the parameters to be optimized in the SVM model are optimized by using the combination of grid search method and standard particle swarm algorithm,and then the hand muscles are optimized in three states of excellent,good and poor The classification of the hand muscles is then carried out in three states: excellent,good and poor.The experimental results show that the proposed method has the highest average recognition rate for hand muscle motor ability classification.3.In order to further optimize the evaluation method in the existing hand rehabilitation training system,a virtual hand rehabilitation system based on Leap Motion and Unity3 D was designed and implemented.The system adopts the Leap Motion sensor,which is popular nowadays,and conducts an in-depth study on the semantic recognition of hand gestures in the system by combining the rehabilitation gestures of cardiovascular and cerebrovascular diseases;three virtual rehabilitation games of different difficulties are designed in Unity3 D according to the hand muscle motor ability classification of patients,and the reliability and validity of the system are verified by two evaluation index experiments.