Research On Emotion Recognition Of Voice And Facial Expressions For The Elderly

Affective computing is to give the ability to recognize,understand,express and adapt to human emotions for computers,so as to achieve efficient and friendly human-computer interaction.The emotion of human has many kinds of carriers,such as facial expression,speech,physiological signal,and so on.The research on multimodal emotion recognition can promote the development of affective computing.The paper is unfolding research from the establishment of multimodal emotion corpus,feature extraction method of speech emotion,feature extraction method of facial expression and recognition effect of emotion feature and other aspects.The main work and research achievements are summarized as follows:1.In view of the current situation that the lack of emotional database for the elderly and the single modal of emotion database,this paper constructs a video emotion database,a voice emotion database and a image database of facial expression on the basis of the TV serial which goes by the name of "empty nest grandfather",and expounds the method of construction and process of multimodal emotion database.The experimental results show that the construction of multimodal emotion database is reasonable and effective2.In view of the lower rate of emotional recognition for commonly used speech features,In this paper,the sixth level wavelet packet coefficient model is proposed for speech feature extraction and emotion recognition for the elderl-y.extracting the feature of 6 layers of wavelet packet coefficients(Wavelet Packet C-oefficients,WPC),dynamic features(first order difference and two order difference)and global features(maximum value,minimum value,mean value,median and vari-ance)based on the sixth layer wavelet packet coefficient model,amounting to 5760 c-haracteristic parameters.The method of principal component analysis is used to reduce the feature dimension,and the support vector machine is selected as the classifier.By comparing with the Mel Coefficients(Mel-Frequency Cepstral Coefficients,MFCC)and the Fourier coefficients(Fourier parameter,FP),The experimental results show that the speech emotion recognition rate of the WPC feature model is h-igher than that of the MFCC and FP,and the recognition rate of the WPC+M-FCC feature set is also the highest after the feature fusion.It shows that the method proposed in this paper is effective.3.In view of the fact that there are not many researches on facial expression recognition for the elderly,In this paper,a two-dimensional Gabor filter model is proposed for facial expression feature extraction and emotion recogniti-on for the elderly.On the basis of the face image library of empty nest grandfatherr,the two-dimensional Gabor filter is used to extract the local features of the face image accurately.In this paper,40 filters with 5 different scales and 8 different direction s are used to perform convolution operations to obtain the features of images in different positions,scales and directions.Since the feature dimension of the extracted image is higher,it is necessary to reduce the dimension of the feature.The proposed metho-d of dimensionality reduction is.first,the image size is reduced and the number of pi-xels is reduced before filtering.Then,some features are extracted forcibly in the filte-ring process.This method greatly reduces the feature dimension,and the later experiments have proved that the key expression information has not been lost,and the facia-l expression can still be well classified better.Finally,this paper uses the method of c-lassifier is neural network of the multiple decision,similar to multi expert decision of the Adaboost classification,and tries to make the classifier output multiple decisionn,weights or values for each decision have different sizes,in order to distinguish the biggest decision and second biggest decision.The experimental results show that our method is effective for facial expression recognition for the elderly,and h-as certain research value.