Research And Implementation Of Emotional Awareness System Based On Video Information

Computer vision has always been a very hot subject,and the emotional perception has always been the focus,and emotional perception is the same as emotion recognition.With the development of deep learning,many advanced theories and techniques have been put forward,and the effect of face recognition is becoming more obvious.At present,the main direction of face emotion recognition is based on static picture or video.This paper is based on video recognition.Video has larger volume and more influence factors then static pictures.Therefore,deep learning method is effective for learning prediction.In this paper,we build a deep learning model,and then build a video emotion recognition system based on Web service through the trained model.The main contents of this paper are as follows:(1)The introduction of deep learning network used in this paper.Elaborate the related theoretical knowledge of deep learning model,from BP algorithm to neural network,from basic neural network to convolution neural network,recurrent neural network,3D convolution neural network and LSTM(Long Short-Term Memory).Analyze the application scenarios and shortcomings,and introduce some important concepts such as the down sampling layer,kernel function,variant models and memory cells of LSTM.(2)Separates the video stream to get the picture frame sequence and audio information.Extract the valid data by face detection module and filtering module,using CNN-RNN to build a multi network learning model,on the other hand,to construct a single C3 D model.On the basis of original network,the emotion recognition of face detection and audio is improved.Finally,the trained model was carried out on the AFEW database,and the accuracy rate of 58.91% was obtained.(3)Finally,build a video stream emotion detection system based on Web service.Introduce the functions of the system process and related modules,including file upload,audio separation,face detection and other back end function modules.Provide web pages for user access and display of detection results.Implement the message queue and describe the structure,provided the asynchronous processing capability and decoupling ability.Provided speed regulation and priority queuing to improving the efficiency of the system.Test the system on the existing database or the data collected by the camera.