Research On Human Lower Limb Motion Intention Recognition Technology Based On Multi-source Information Fusion

Exoskeleton as an intelligent device to assist the human body is widely used in medical rehabilitation,civil and military applications.The premise of exoskeleton softly realizing human body assistance is to be able to accurately perceive the movement intention of human lower limbs,that is,to first recognize the current human motion pattern,and then recognize the gait phase in this pattern,so as to make corresponding movements to assist human movement.To address the current problems of exoskeleton motion pattern and gait phase recognition,the following work has been accomplished in this paper:(1)Design of lower limb motion information acquisition system.By analyzing the human lower limb motion mechanism,a lower limb motion information acquisition system based on myoelectric sensors and IMU(Inertial Measurement Unit,IMU)is built.In the experiments of motion pattern and gait phase recognition,the feature recognition accuracy based on the fusion of multi-source information data both reached over 93%,which is better than that of single-source information data.(2)Gait feature extraction based on multi-source information fusion.The algorithm of Dual Stream CNN-Relief F is proposed to extract gait features by combining two algorithms of Dual Stream CNN(Convolution Neural Network,CNN)and Relief F.In the motion pattern and gait phase recognition experiments,the recognition accuracy of the extracted features of this algorithm is more than 96%,which is higher than the empirical formula method.(3)Based on the ISSA(Improved Sparrow Search Algorithm,ISSA)optimized SVM(Support Vector Machine,SVM)motion pattern recognition.The ISSA algorithm is proposed to optimize the hyperparameters of SVM.The ISSA-SVM algorithm achieves 99.30% recognition accuracy for four motion patterns,which is2.80%,2.55%,and 1.46% higher than the recognition accuracy of SVM optimized by other three algorithms,respectively.(4)Research on gait phase recognition of multi-motion patterns.Based on the characteristics of different motion patterns and the actual control requirements,the gait phases of four motion patterns are classified.The gait phase recognition algorithm for multi-motion patterns is investigated,and the accuracy of LSTM(Long Short Term Memory Networks,LSTM)or gait phase recognition of all four patterns is over 98%in comparison.In summary,the proposed lower limb motion information acquisition system and Dual Stream CNN-Relief F feature extraction algorithm provide a favorable data base for the recognition of human motion intent by exoskeleton;the proposed ISSA-SVM and the selected LSTM algorithm can accurately recognize different human motion intent respectively.