Design And Implementation Of Digital Facial Expression Real-time Recognition System

With the popularity of intelligent devices,human-machine interaction and machine vision have become a research hotspot.Machine vision system refers to converting the captured object into image signal through the machine image capturing device and transmitting it to the special image processing system to obtain the morphological information of the captured object.Photoelectric image processing technology converts analog image signal in nature into digital image in computer,and makes the machine have the same visual function as human beings through related artificial intelligence algorithm.Facial expression recognition,as an important interface of human-machine interaction,has very high application value and prospect.Nowadays,image recognition technology based on in-depth learning has developed rapidly,especially convolution network has become the mainstream network architecture of image recognition.This paper mainly studies the relationship between network size,network structure and recognition rate in facial expression recognition,and proposes an improved scheme.The effect of the lightweight neural network model in the field of expression classification is studied,and the recognition rate is kept weakly reduced while reducing the complexity of the model.Use the method of expression label distribution to solve the problem of composite expressions.Finally,the expression recognition analysis system is designed and implemented,and deployed on the embedded platform.The results show that the deep recognition-based expression recognition algorithm can achieve real-time effects on small devices,and the key to deep learning algorithms can be deployed in realworld scenarios.step.The main work of this paper is as follows:1.In order to make the network model designed in this paper have good practicability in complex scenes,FER2013 expression recognition database is selected as the experimental data set.The shallow network model AlexNet and the deep network model VGGNet are analyzed and studied.In order to solve the problem that the two models can not take into account both model size and recognition performance,a lightweight network model is designed to reduce the scale of the model while ensuring that the recognition rate is not lost.The scale regulation factor is added,and the balance between network size and recognition performance is explored.In order to solve the problem of multiple facial expressions at the same time,the label distribution based learning method is used.Compared with the conventional single expression data set method,the recognition rate is greatly improved.2.Based on the lightweight model designed in this paper,a simple expression recognition and analysis system is designed,which includes image display,face detection,prediction category,feature visualization and other functional modules.Finally,the model is transplanted to the embedded platform,and the real-time recognition speed of VGGnet and lightweight network on the embedded platform is compared.Experiments show that the proposed lightweight model achieves the basic requirements of real-time.