Design And Implementation Of Facial Expression Recognition System Based On Patrol Service Robot

With the development of high-tech technology,products combing "artificial intelligence" and"Internet +" are popular.Product functions are essential,the interactive fearure of the product are more attractive for customers.At present,the exhibition hall security patrol service robot has a good future.It is attractive to improve the robot interactive capability.Based on the actual need of project,the paper develops a facial expression recognition system for patrol service robot,with the help of OpenCV visual open source library.This paper states the research situation of current facial expression recognition problem.The process of expression recognition includes face detection,feature extraction and classification.Comparing the advantages and disadvantages of classic expression recognition methods,the paper selects the methods for this project.The main work of this paper can be summarized as follows:1)Research on Window Setting Method and Extraction Strategy of Gabor Filtering.This paper focuses on the influence of different Gabor filter window setting methods and sampling strategies on the recognition performance and operation efficiency of the system.Through the deep analysis of Gabor filter algorithm,the window setting method and local sampling strategy are experimented.After analyzing the experimental data,observe the experimental results,select the optimal window setting method and feature extraction strategy.2)Construct SVM classifier and realize the expression recognition function.Based on the idea of bilinear method and grid search method,the radial basis function model is designed,and the optimal parameters of radial basis SVM classifier are found by experimental data.3)Construct Back Propagation artificial neural network model(BP ANN)for the facial expression recognition.This paper chooses the relevant parameters and compute the number of node in each layer and hidden layer levels to construct the BP ANN of the system.By comparing the recognition rate of the sigmoid function and the tangent function,sigmoid function is chosen as the activation function of the system.4)Create custom training samples for the system.By using JAFFE,YALE and CK + open source sample libraries,The system can make change to the samples or gain new samples needed.5)Complete the development of facial expression recognition system for patrol service robot,passing case test.According to the software engineering development process,combined with the theory of expression recognition,the system has designed and developed based on OpenC V library.The classification and recognition performance of the system using SVM classifier and BP ANN were evaluated by plotting the "precision-recall rate" curve.Test results show that the training time of SVM classifier is is far less than the BP ANN,though BP ANN has better recognition performance and classification for expression recognition.In above two kinds of classifier,for happy,anger,surprise of recognition rate higher than fear,disgust and sadness.