Research Of Real-time Gesture Recognition System Based On Dual-channel SEMG Signals

All life processes in living organisms generate electrical signals,which we call bioelectricity.The generation of bioelectrical signals is the constant change of membrane potential caused by the stimulation of biological cells.It can spread rapidly through the body,generating different potentials at different locations on the body’s surface.It is the existence of this potential difference that enables bioelectrical signals to be detected and analyzed.The generation of complex bioelectricity is closely related to the life state and contains rich physiological and pathological information,so it is widely used in the fields of disease diagnosis,condition monitoring and rehabilitation training.Common and widely used bioelectrical signals are EEG,EMG and ECG.Among them,as a non-invasive means of EMG signal acquisition,surface electromyography(s EMG)signal has been widely concerned in academia and industry,and is the key technology of non-invasive human-computer interface,which has a good application prospect in the field of human-computer interaction.s EMG signal has been widely used in gesture recognition,prosthetic control,muscle fatigue detection and other fields.There are mainly two acquisition methods for s EMG signal detection:multi-channel detection and high-density detection.The current technology of multichannel detection accuracy is low;However,high density detection requires more computation.In order to better apply to ordinary users,it is necessary to build a gesture recognition system with high accuracy,good real-time performance,little computation and simple operation.This paper proposes and implements a multi-channel real-time gesture recognition system based on Open BCI and Python.The system can collect,process and analyze s EMG signals in real time.Data processing and analysis include data cleaning,data preprocessing,feature extraction and classifier recognition.In addition,a user interface(GUI)program is designed to realize real-time visualization of data and real-time feedback of recognition results.The feasibility of the identification system is verified by practical test.This paper combines the concept of static image in the field of image recognition and designs the static data as the data collected by the sensor when maintaining the final motion of the hand.A method of matching real-time data with static data,namely static recognition,is proposed,which avoids the mark of switching gesture.Finally,a gesture recognition algorithm based on one dimensional convolutional neural network(Conv1D-Ind RNN)model is proposed to further improve the accuracy of the real-time gesture recognition system.The feasibility of the proposed algorithm is verified by open-source data sets and data collected by Open BCI.Compared with the traditional s EMG gesture recognition system,this system has the advantages of low hardware cost,compact and efficient algorithm,and good real-time performance.The main research contents and innovations of this paper are as follows:1.Construction of s EMG Dual Channel Real-time Gesture Recognition System Based on Open BCI and Python Language.The PC GUI program of the dual channel real-time identification system based on Open BCI and Python is developed.PC and Open BCI hardware are connected through Bluetooth,and GUI program realizes the functions of receiving,processing,real-time display,storage,classification and recognition of Bluetooth data.Design the random forest,support vector machine(SVM),artificial neural network,a one-dimensional Convolutional Neural Networks and Convolution Neural Network-Gate Recurrent Unit(Conv1D-GRU)five models,respectively,the dual channel,three kinds of gestures and its Root Mean Square,the Mean Absolute Value,Variance,Simple Square Integral,sum of the amplitude spectra of the STFT,Median Frequency characteristics of six kinds of s EMG signal analysis and processing.Among them,the method based on Conv1D-GRU model can achieve99% accuracy and recall rate for the static data of 5 subjects,while the accuracy rate in the real-time test can reach 96.7%.According to statistics,the average update time of the whole GUI is less than 120 ms,which meets the real-time demand.2.A Gesture Recognition Method Based on Conv1D-Ind RNN Network.On the basis of the dual channel real-time identification system based on Open BCI and Python language designed in this paper,the Ind RNN network is introduced and the Conv1 DInd RNN hybrid model is designed.One-layer Conv1 D and two-layer Ind RNN are used to extract intra-channel and inter-channel features respectively,and the spatial and temporal features of time series signals are combined.This model evaluates 18 kinds of gesture signals from 10 subjects in the open source Ninapro subdataset DB5.Using only one feature of root mean square and 10 Epochs,the model achieves the same effect as the latest model in the same dataset,with an accuracy of 87.43%.More importantly,the number of parameters in our model is only 5278,which is far lower than the existing model.Then the model was applied to the two-channel real-time recognition system established in this paper,and the accuracy rate of 10 subjects reached 99.1%.3.Verify the feasibility of the system and model.Five testers were recruited to verify the two-channel real-time recognition system based on Open BCI and Python language proposed in this paper.Three basic gestures,two channels of differential signal input and classical machine learning classification algorithm were adopted to verify the system.The experimental results proved the feasibility of the system.The feasibility and recognition effect of the Conv1D-Ind RNN hybrid model were verified by open source data set and Open BCI identification experiments,and the robustness of the model was proved.