Research On Face Expression Recognition Based On Video

Facial expression recognition is one of the research hotspots of human-computer interaction and pattern recognition.The application scenarios of facial expression recognition mainly include human-computer interaction,intelligent control,safe driving,medical treatment,communication and other fields.Today' s facial expression recognition research mainly uses static images as the research object,but static images often lose many dynamic characteristics,so often it is impossible to accurately capture some micro-expressions of the object.Based on this,this paper will dynamically capture the subtle changes in expressions through the analysis of video data,and better reflect the real expression features of the object.The main research work and innovations of this paper are as follows:1.For static pictures,use VGG convolutional network structure,reduce the parameters of the network by fine-tuning the parameters and the topology of the network,so that it can run on the low-end configuration computer,and keep the accuracy and robustness of the algorithm as much as possible.2.For the image sequence,a new method of extracting the avatar frame is proposed firstly,and then the idea of dual-stream convolutional neural network is adopted to construct a dual-flow network.The spatial flow network uses the static image trained VGG network,time flow.The innovative 3D convolutional network structure in the network inputs the optical flow graph of adjacent frames and then performs fusion training.After the simulation test of the data set,it is found that the model can better complete the classification of various dynamic expressions.