Face Expression Recognition And Analysis System Based On Machine Learning

Face expression recognition is an important research topic in the fields of computer vision and pattern recognition,and it is an important part of intelligent human-computer interaction technology.Studies have shown that facial expressions play an important role in human emotional communication,and people can get a lot of information from facial expressions,such as gender,mood,health status and so on.Intelligent recognition of facial expressions is beneficial to monitor human health in smart home environment,which is a hot topic in the field of AI application research.In this thesis,machine learning algorithms are used to realize real-time detection of face targets in images or videos and to realize classification and recognition of face expressions.Machine learning algorithms are also used to explore the optimal methods of existing face detection and expression recognition algorithm models,and various functional modules are transplanted to embedded platforms to build intelligent facial expression recognition system on the mobile side.The specific research contents include the following points:A.Studying and realizing a variety of face detection algorithms,comprehensively comparing the performances of different algorithms,and considering the limited hardware computing capacity after inletting algorithm to the embedded end,MTCNN,which is of good detection effect and real-time performance and based on convolutional neural network,is selected as the basic algorithm of face detection and be improved and optimized.A improved face detection algorithm Res-MTCNN is proposed to have a better accuracy rate in the premise of satisfying the calculation ability;B.Face expression recognition belongs to the classification problem in the field of image recognition.To meet the task needs of the expression recognition module,Deep ID network model based on deep learning in the field of face recognition is selected as a basic model of expression recognition function.And the excellent face feature extraction ability of the original network was utilized,meanwhile,residual module and attention mechanism were introduced,so that themodified network model can be more adapted to the expression multi-classification recognition task,the accuracy rate of the expression classification task proposed in this system reaches 96.5%;C.In terms of program operation,it is proposed that the two-threaded collaborative loop work makes face detection and expression recognition run in parallel,that is,face detection of the next frame is immediately conducted when expression recognition for the previous frame in video data is still recognizing.This way of operation of this program can not only improve the time utilization rate of the system,but also provide more available detection time for face detection module to improve the detection accuracy of the model;D.The NVIDIA Jetson TX2 platform deployment system with GPU chip is selected,and the facial expression recognition system has been implemented on the embedded mobile side.Experiments show that in this work,real-time recognition and analysis of the video are completed based on the proposed human face expression recognition and analysis system,and each part of the function module on the server side can be completely transplanted to the Jetson TX2 embedded system,which can carry out the real-time recognition and analysis of the video facial expression.Compared with the operation results of GPU server side,the recognition accuracy of the embedded end with the functional modules has been maintained,and the operating speed has reduced to a quarter of the original speed while meets the real-time needs.As a result,the successful deployment of the system on the embedded end provides a feasible solution and useful reference for the realization of new human-computer interaction and presents potential application values in the future.