Design Of A Rehabilitation Robot With Behavior Recognition

To meet caused by diseases such as cerebral apoplexy,hemiplegia patients with lower limb trouble walking on rehabilitation training needs,solve the rehabilitation training in patients with poor equipment,physicians matching the problem of insufficient,carried out the mobile lower limb rehabilitation training robot mechanism design and the development of control system,at the same time,for the patients to combine practical training condition,In the process of rehabilitation training,the corresponding training mode is formulated and the multi-sensor fusion behavior recognition system is built,which provides the basis for the classification,recognition,decision-making and adjustment of training mode of patients’ various behaviors and movements in rehabilitation training.The research contents of this paper mainly include:Accurate identification of behavior state can provide a judgment basis for the formulation and switch of rehabilitation strategy of rehabilitation robot,and at the same time provide necessary control input for the underlying controller.According to axiomatic design theory,a multi-sensor information collection scheme is determined and a multi-sensor fusion behavior identification system is built.Behavioral information during rehabilitation training is collected from the aspects of spatial distribution,dynamics and kinematics in an all-round and multi-dimensional way to provide simple and comprehensive feature input data for the behavior recognition algorithm.Combined with the two-level behavior recognition network,the behavior state of users in the rehabilitation training process is classified and identified.The first level of the second level behavior recognition network is probabilistic neural network,which is mainly used to judge whether patients have fallen.The second level behavior recognition network chooses SVM network,and MATLAB/Simulink is used to simulate the controller.The performance of behavior recognition and classification is compared by comparing BP neural network with SVM support vector machine and its kernel function.The control strategy with higher accuracy and best performance was selected as the classification algorithm of behavior state recognition.Network in MATLAB platform to build the secondary behavior recognition,the acquisition of 300 groups of data feature extraction,the normalized processing,after the reorganization of the data,the data input to the network behavior classification and recognition,210 groups of data before using the training model of PNN classification and recognition,in another 90 sets of data to validate the model effect of classification and recognition,classification comparison results show that theThe accuracy is 100% and can be used as the first layer of fall recognition network.In order to solve the specific classification and recognition problems of various behaviors in the process of rehabilitation training,the second layer uses SVM support vector machine to compare with the traditional classification method BP neural network.The classification and recognition accuracy of BP neural network is 80.64%,and the simulation test is conducted on the four kernel functions of SVM.The research shows that the accuracy of SVM-RBF network is close to 98%.Finally,RBF is selected as the kernel function of the second layer SVM behavior recognition network,which can meet the requirements of safety and applicability in the process of rehabilitation training.In the process of training,the movement control performance of rehabilitation robot is stable,which can meet the requirements of rehabilitation training speed and weight loss of hospital patients.Finally,the experimental verification shows that the fall recognition performance of the first-level behavior classification and recognition network is still 100% after the training of the two-level behavior recognition network model based on the training data collected and processed by the multi-sensor information acquisition system,and the classification accuracy of the second-level behavior classification and recognition network for the 12 kinds of behaviors in the training process can reach 96.7%.It is shown that the multi-sensor fusion gait recognition and classification system is reasonable in design and feasible in method,which can better realize gait recognition and classification,and has strong robustness against external interference.