Face Tracking And Facial Expression Recognition Based On Deep Learning

Facial expression is an important part of the body language,it is a reflection of human physiology and psychological activity which can convey the inner emotional state of the human.Recently,facial expression analysis has become a very hot research topic in the fields of computer vision,pattern recognition,and artificial intelligence.Besides,Deep Learning has developed rapidly in recent years and has been widely used in the research of computer vision which achieves pretty good performance.This paper aims to use deep learning methods to study the two key technologies of face tracking and facial expression recognition which are involved in facial expression analysis.On the one hand,face tracking is a very important forefront technology in the research of facial expression analysis,and it is a prerequisite for developing a real-time facial expression analysis system.On the other hand,facial expression recognition is the ultimate goal in facial expression analysis.Actually,an advanced facial expression recognition technology has great application value.The contribution of this paper is summarized below:(1)This paper studys the face tracking problem in real scenes.In order to improve the accuracy of face tracking,a face tracking algorithm based on spectral filter is proposed,which uses the properties of rotation invariance and translation invariance of graph.The algorithm is not only simple and effective but also easy to implement.Futhermore,it can well cope with the often occurring deformation and rotation problem in the face tracking process.The proposed algorithm achieves very good results both on the self-built child face tracking dataset and standard object tracking dataset.(2)This paper researches the dynamic expression recognition based on the real scenes in videos.A dynamic expression recognition algorithm based on convolutional neural network and recurrent neural network is proposed.This method makes full use of the ability of deep convolutional neural network to extract spatial features and the ability of recurrent neural network to fully explore temporal context information.The proposed algorithm can not only model dynamic expression but also can extract the robust temporal and spatial features of facial expressions in videos which is able to classify dynamic expression effectively.(3)To better research the problem of children's smile detection,a smile detection model based on convolutional neural network named Smile-Net is designed.The Smile-Net is a light-weight network that has less than layers and parameters compared with the traditional networks,but it can achieve better results than the methods of the classical deep neural network.Besides,this paper develops a children's face tracking and smile detection system.The system can analyse the video of children who participate in the parent-child interaction experiments accurately.The generated smile results can be used as an auxiliary analysis method for evaluating children's social emotional competence.