Research On The Application Of Face Recognition And Expression Recognition In Human-computer Interaction

Today,the communication between people and machines is more frequently.Traditional human-computer interaction modes have problems such as inconvenience,not ergonomic,and low efficiency.With the emergence of intelligent robots,virtual reality and unmanned scenes,the traditional human-computer interaction can no longer meet people's needs,and natural,active,and rich information become the characteristics of intelligent human-computer interaction.Recent years,with the rapid development of computer vision,researchers begin to be interested in human-computer interaction based on vision.Cameras,as the "eyes" of machines,capture information,gets rid of the traditional interaction modes of command lines,texts and graphics,and gradually promote the transformation from human-computer to "human-human".As the most direct way to express identity and emotional information,face can efficiently transmit a lot of information in visual human-computer interaction.However,the technology of face-based visual interaction also relies on the development of face and expression recognition.At present,benefit from big data and deep learning,face recognition has developed rapidly,but there are still great challenges in how to maintain high-accuracy recognition in complex real scenes.At the same time,face recognition based on deep learning relies on huge datasets and complex feature networks,how to compress the models and improve real-time performance also need further study.In addition,compared with facial recognition expressing single identity information,expression recognition expresses rich emotional information,which is more important in the application of visual human-computer interaction.However,due to the low quality of the public datasets,the lack of expression categories,and being susceptible to irrelevant information in the images and so on,there are greater challenges in expression recognition.To solve the problems above,the specific works are as follows:(1)This paper proposes a face recognition system combined with model compression,which uses MTCNN model for face detection.After preprocessing,the face image is sent to the feature network based on ResNet,and the feature vectors are embedding into Euclidean space for face recognition.The network is trained by the triplet loss function.In order to solve the problem that the triplet loss is hard to be trained,a joint training method of SoftMax loss and the triplet loss is proposed,and the selection method of triplet pairs is optimized.In order to compress the model and improve the real-time performance,a method of model pruning is used in the feature model,and the ResNet with multi-branches is optimized.Theoretical and experimental results show that the proposed method has high accuracy and good robustness,and has miniaturized model,real-time recognition,so it can be better used in human-computer interaction.(2)In order to improve the accuracy of facial expression recognition and enrich the information in human-computer interaction,this paper further proposes a multi-class expression recognition model based on multi-layer feature fusion.The network has a lightweight structure,which efficiently fuses the shallow local features and deep global features of the network,and determines the appropriate parameters through experiments.Aiming at the problem of low quality datasets,this paper makes data balance and augmentation,enriches the data and improves the data quality.During inference,the input data is also expanded,and the average result is taken as the output.In addition,in order to make the expression more separable and make the information more accurate and rich,this paper increases the expression classification from the traditional 7 categories to 9 categories by building and marking the self-built dataset,and verifies its rationality.Theory and experiments show that the facial expression recognition algorithm proposed in this paper can classify 9 kinds of facial expressions in real time with good accuracy.(3)Taking the face recognition and expression recognition algorithms proposed in this paper as the core,a friendly interactive interface is designed.The expression information is fed back quantitatively and qualitatively through the visual interaction in the natural condition which is used for the expression training to art performers.Then the interface and functions are displayed,and the characteristics of natural interaction around the intelligence interaction are evaluated,to further verify the practical application value of the algorithm in this paper.