Application Of Emotion Analysis System Based On Video

Face expression recognition has always been a major part of emotional research,and it is also an important part of future social human-computer interaction.It has very important research significance and broader application scenarios,such as detecting fatigue driving,mental health,distance education and other fields.In the development process of facial expression research and recognition,two models can be constructed according to the research method: the traditional scheme based on classifier and end-to-end deep learning model facial expression recognition is mainly divided into four steps: the face in the image Detection,face key location,expression feature extraction and expression recognition are the four main steps.The first two steps have been studied in many fields,but the key point location and feature extraction of the face are still the core of the research.The above mainly describes two main research methods.When we consider the objects to be studied,there are roughly two types,based on static images and facial expression recognition based on real-time video sequences.In order to solve the analysis of facial expressions from static images,we propose a neural network model based on isolation loss to solve.The main feature is to use convolutional neural networks to obtain facial expressions from images.Compared with the artificially designed Haar feature and its LBP feature,it has better accuracy and rationality.In addition,based on the central loss,this paper proposes a new loss function-isolation loss,which effectively improves the discriminability of the system and solves the misjudgment.In the processing of video sequences,we propose a local bidirectional recursive cyclic neural network to solve this problem.The main principle is to input various parts of the human face in the two-way cyclic neural network model,and then extract the time-based change information of each part and input it into the high-level network fusion,and finally get the expression change of the whole face area at different times..Thereby classifying and recognizing facial expressions.In order to further improve the expression recognition rate in real-time video,we also combine the time series information and spatial information to carry out the final model fusion,so as to better perform the expression recognition in the video sequence.The work done during the entire study was tested on different data sets.In the process of face emotion recognition based on static images,we carried out experiments on the dataset of FER-2013.The isolated loss method studied effectively improved the recognition rate and robustness of facial expressions.In the expression recognition of video sequences,we tested on the CK+,Oulu_CASIA and MMI data sets.The final experimental results show that the PHRNN network model has achieved a good recognition effect,and more abundant information is extracted compared to other models.The final expression prediction using this model effectively improves the accuracy of recognition.