Design And Implementation Of Facial Expression Recognition System Based On ARM

Facial expression recognition is the recognition of human facial expression,which separates the specific expression state from the static image or dynamic video sequence to determine the psychological emotion of the identified object.Through various methods,facial expression recognition technology is to enable computers to have the ability to understand and express,just like as human beings,to adapt to the environment independently and to give human feedback to achieve the goal of human-computer interaction.Expression recognition is widely used,and it has great scientific research and practical value in medical,education,driving,and games.In deep learning,the convolutional neural network is a hot topic and technology in the field of artificial intelligence.It has achieved remarkable results in the fields of speech recognition and image recognition.The current research on expression recognition technology mainly focuses on updating algorithms and expression databases,and does not focus on the technical implementation of mobile terminals.In view of the wide application of expression recognition,this thesis will use face detection technology and related convolutional neural network of deep learning,combined with high-performance PC(server)to complete the deployment and implementation of expression recognition system on embedded mobile terminals.According to the main research contents of this paper,the design and implementation process of facial expression recognition system and related technical knowledge are expounded in detail.On the basis of theoretical knowledge,the concepts related to neural networks and the related technologies and methods are introduced.In this design,the expression recognition system is inseparable from the neural network,and the neural network can't be separated from the expression picture data.The current expression database is relatively large,but the open source database is rare.The total amount of data that can be downloaded from expression database is also slightly insufficient when used in a neural network.Therefore,I will deal with four databases separately and merge them into large-scale facial expression data.This database contains about 200,000 facial images,which will be used for the training of facial expression recognition models.Facial expression recognition is the same as the basic steps of face recognition.It is necessary to detect a face in a given picture or a sequence of moving images and preprocess the corresponding image.In this paper,the face detection technology of normalized pixel difference is improved by adding a limit of the facial detection window and detection scoring system,which makes the algorithm of detection more efficient,and the detection speed is faster,which is more suitable for low-performance ARM terminal processing equipment.In order to be suitable for embedded devices,based on the research and analysis of neural network theory and various models,this paper chooses a new lightweight network model structure-MobileNet to train the expression recognition model.After comparing and testing the influence of two factors on the classification results of the network,a group of factors that perform well are selected as parameters,and Caffe is used to complete the training of the network model on high-performance PC.This model not only reduces weights and other parameters by an order of magnitude,but also improves the classification speed.This thesis uses the Fer2013 test set picture and the real-world expressional pictures to test the accuracy of the model.The model achieved an average recognition accuracy of 77.4% in the Fer2013 test set and 63.6% in the real environment.Considering the cost-effectiveness,the system uses the Raspberry Pi as the core processor of the embedded terminal.Due to performance limitations,the Raspberry Pi cannot independently complete the process from detection to identification.Therefore,I will use a high-performance PC as a server to design the entire system.The system collects the emoticon image through using the camera of Raspberry Pi,and sends the emoticon image to the ARM to detect face.The detected face will be transmitted to the server through TCP/IP,the server will recognize the facial expressions and return the results.The ARM board will use the QT interface to display the currently recognized image and expression recognition results on the monitor.Through continuously testing and optimizing,the function of the facial expression recognition system has been improved,and the recognition speed of the system has been significantly improved.