Research On Eye Movement Direction Identification Method Based On EOG Signal

In the traditional field of human-computer interaction,human gestures and limb movements are used to express human intentions,and computer control is completed.However,for people with limb disorders,it is impossible to complete the operation.In order to solve the problem of human-computer interaction,bioelectric signals are widely used to reflect human intentions.This topic is to convert eye movement into electrical signals to achieve the purpose of human intent to manipulate external mechanical equipment.Electro-oculogram(EOG)is a relatively weak bioelectric signal,which is vulnerable to interference from the external environment.Because of the interference of invalid and meaningless EOG signals such as unconscious blinking,the accuracy of the EOG signal is poor and the recognition rate is low.Eye movement signal recognition rate is high,which makes up for the problem of low recognition rate of EOG signal.In this thesis,a method of training by eye tracker is proposed.When collecting the EOG signal data,the Eye Track is tracked by eye tracker to remove the meaningless EOG signal such as invalid eye record and unconscious blinking.After eye movement training,the accuracy of training set of EOG signal has been greatly improved.Then the EOG signal is pretreated by wavelet threshold de-noising,normalization and endpoint detection.Secondly,feature extraction of denoised EOG signal is carried out in time domain,frequency domain and time-frequency domain,and the relationship between each feature value and different eyes is analyzed quantitatively.The time-frequency characteristic based on wavelet analysis has a certain relationship with EOG signal,and through the peak detection method,we can see that there are great differences among different kinds of EOG signals by setting threshold.Finally,in the process of pattern recognition,three classifiers,namely BP neural network,linear discriminant analysis(LDA)and support vector machine(SVM),are selected to input the extracted eigenvectors in each classifier.Five different kinds of EOG signals are used as action recognition methods to classify and recognize.By comparing and analyzing the classification results,the recognition rate of EOG signal trained by Eye Tracker is higher than that of untrained,and compared with the recognition results of three classifiers,the recognition efficiency of SVM is better.