| Human-computer interaction technology proceeds from human-aided-computer to computer-aided-human mode.Currently,the new human-computer interaction mode is performed based on bioelectrical signals in which computer needs to understand the user's intention and make corresponding response.Bioelectrical signals that can be used for human-computer interaction include:electrocardiograph(ECG),Electroencephalogram(EEG)and Electromyography(EMG).EMG signal is easy to be recorded,with high signal-to-noise ratio,rich modes and high accuracy,which makes it attracted more and more attention in human-computer interaction.Among them,hand gesture recognition based on arm EMG has been widely studid.Compared with gesture recognition,mouth-shapes recognition based on facial EMG is more,suitable for silent interaction with rich commands which would be more conventient for deep paralyzed people.However,the related research is still limited,for instance,the lacking of mouth-shapes recognition algorithms and prototype system development.This paper mainly studies the key technologies of human-computer interaction based on mouth-shapes.The mouth-shapes is generated by the traction of many muscles.Different mouth-shapes will cause the muscles to participate in different degrees,resulting in many different patterns of EMG signals in energy,frequency and location.Therefore,we can capture the action potentials of different muscles through multiple electrodes,and distinguish the EMG patterns of different mouth-shapes using time-series processing and machine learning techniques.Then the EMG mode caused by different mouth-shapes is used as the control command output to drive the machine to make corresponding response,so as to achieve the purpose of human-computer interaction.In this paper,based on the facial EMG signal,we can accurately recognize various mouth-shapes.The main works include:1)The research of vowel mouth-shapes recognition algorithm based on face EMG is studied.The EMG data of 5 Chinese vowel patterns were collected from 10 subjects.Through a large number of comparative tests and the proposed algorithm of feature evaluation,the optimal algorithm of pattern recognition is determined.The results show that using five channel data,extracting six best features and combiningwith linear kernel support vector machine has the best recognition performance,and the average classification accuracy is 96.5%.The method of data enhancement is used to solve the problem that the model can be used across subjects,so that new users only need a small amount of training to achieve high classification accuracy.We also analyzed the possibility of increasing the number of orders and combining the vowel mouth-shapes to enrich the commands.Finally,the algorithm is applied to the online system and the average accuracy is 92.6%.2)Numerical lip shapes recognition algorithm based on facial EMG is explored.In the experiment,we collected the EMG data of 25 subjects' English pronunciation mouth-shapes of "Zero" to "Nine".Through many comparative experiments,a new feature extraction method is proposed which is combined with the second-order polynomial kernel support vector machine.Finally,the average classification accuracy of 10 numerical mouth-shape recognition of all subjects is 93.2%.3)Combination of feature migration,and data enhancement,which solves the problem of using 10 kinds of digital mouth models across subjects,so that new users only need a little training,to achieve high accuracy.The combination coding method of digital mouth-shapes is proposed,which can produce enough commands to deal with complex human-computer interaction scenarios.4)A human-computer interaction system based on face EMG is designed and implemented.Different interaction modes are suitable for different scene requirements.The human-computer interfaces for diverse interaction scenarios have been developed,such as robot arm,digital dialing,typing system,etc.,which have achieved good performance and proved the effectiveness and stability of the algorithm. |
PDF Full Text Request