The Research On Service Robot Intelligent Interaction System Based On Facial Expression Recognition

With the rapid development of intelligent robots,how to give robots harmonious human-robot interaction ability makes it able to perceive human emotion has been becoming the current hot topics in the study of human-computer interaction.Facial expression as an important part of human emotional expression,to enable the robot to understand the emotional expression of human,we must make the robot have the ability to recognize facial expressions.Therefore,the study of facial expression in the field of human-computer interaction has shown its important research significance and application value.According to the problem of insufficient representation of facial expression features and the poor performance of real-time feature extraction using traditional AAM model,a novel extracted shape and texture features of facial expression method based on the combination of BRISK and AAM was proposed.Firstly,the Fast-SIC algorithm was used to fit out the AAM of face for the original face image.In order to enhance feature matching efficiency,BRISK algorithm was used to match the acquired key facial feature points,and then LGBP was used to extract texture features of AAM in order to strengthen the ability of facial expression features representation.Finally,to classify the classes of facial expression features using the SVM classifier.However,the traditional method of facial feature location extraction of facial features,inevitably the introduction of artificial calibration error,and convolution neural network can overcome this shortcoming.in this paper,Tensor Flow,currently popular deep learning framework,was applied to design a unique convolutional neural network model for express ion recognition.The difference between the convolutional layers designed by this paper and the traditional convolutional layers was that we ignore the biases of the convolutional layers,accelerating the training speed while simultaneously reducing the nu mber of parameters to learn.By analyzing and comparing the advantages and disadvantages of the different processing techniques of the network layers(such as activation function,convolutional kernel size,Dropout,etc.),the more suitable network structure and parameters are selected,and the eight basic expression categories were effectively classified.In this paper,the effectiveness of the algorithm of expression recognition was experimented validation on the datasets of CK+ and JAFFE,and the facial expression recognition method combined AAM with LGBP achieved an average of recognition rate of 92.67%,and facial expression recognition method based on convolution neural networks achieved an average of recognition rate of 95.54% in the CK+ dataset.Finally,we design an intelligent interactive system based on facial expression recognition,and validate the effectiveness of the algorithms proposed by this paper on the robot platform of NAO.