Research And Implementation Of Facial Emotion Recognition System Based On Spatio-temporal Two Stream Network

With the advent of the Internet age,the artificial intelligence industry has become a leading industry in the country's economy and technology.Among them,artificial intelligence recognition through facial emotions has become a research hotspot in the current society.According to current research findings,in the process of users communicating with each other,most of the information is conveyed through the medium of expressions.When users express their expressions,their facial expressions will also change accordingly.Therefore,users can use Changes in facial expressions to identify the user's emotional change trend.Emotions can be expressed in many different forms that are invisible to the naked eye.With the right tools,you can detect and recognize the facial emotions of human faces.In the past few years,the demand for human emotion detection is increasing.Emotion recognition began to be widely used in fields such as human-machine interface,animation,medicine,and security.Based on facial emotion recognition,this paper studies and implements a facial emotion recognition system,combines micro-motion and macro-motion functions,and proposes a spatio-temporal two-stream network to improve video emotion recognition.The network integration structure captures information about micro and macro motions that will be beneficial to the prediction of emotions,that is,smaller and shorter micro motions are analyzed through the two-stream network,while larger and longerlasting macro motions can be followed by The recursive network is well captured.For the design of facial emotion recognition system,the system structure is determined to be B/S architecture,and the overall framework,functional structure and overall process of function realization of the system are designed.According to the system process,the related database structure and functions are designed.Before performing emotion recognition on human faces,first preprocess the Aff-Wild data set used in this article,and then transfer the processed data set to the spatio-temporal twostream network for emotion recognition.The spatio-temporal two-stream network proposed in this paper uses a deep learning framework,and the training process is spatio-temporal two-stream parallel operation.In the temporal flow,the inter-frame phase difference is selected to replace the optical flow as the input of the temporal flow,and a series of gray-scale images are fed into the controllable pyramid model.At the same time,the preprocessed RGB image is transferred to the spatial stream improved Res Net50 network for processing,and the ability of facial emotion recognition is improved through the fusion of spatiotemporal features.Based on the system of temporal and spatial two-stream network emotion recognition in this article,the system server is first built,the hardware and software environment required by the system are configured,the various functional modules of the system are improved,and the system is deployed on the local server to test the functionality of the system and stability,and finally successfully debugged the facial emotion recognition system.After testing the facial emotion recognition system,it is verified that the system can well fulfill the requirements in the design.The facial emotion recognition system can upload videos,photos of people on flyers,and multi-person photos from the local site for recognition.It can also perform real-time emotion detection on faces and output the results on the screen.In addition,user information and system information can be managed through the B/S system.After the system has been tested,all modules have reached the desired effect.