Embedded Realization Of A Facial Expression Recognition System Based On Improved LeNet-5

Face expression recognition as the most intuitive way to reflect human emotions is widely used in human-computer interaction,automatic driving,industrial monitoring and other fields.In practical applications,face expression recognition algorithms are often implemented in the cloud.With the vigorous development of Internet of Things technology and the application of the next generation mobile communication technology represented by 5G,more and more edge nodes will access the existing network system.If facial expression recognition algorithm can be implemented in embedded edge nodes,it will greatly reduce the performance burden and maintenance cost of cloud server caused by massive data generated by many nodes,as well as network bandwidth requirements.Under the above background,this topic focuses on the realization of a real-time image acquisition and face recognition system on embedded devices.The main work of this topic includes the following parts:Firstly,the research status of edge computing and facial expression recognition at home and abroad is analyzed,and the related technologies are also discussed.On this basis,the feasibility of the subject is demonstrated.Secondly,facial expression recognition is divided into two aspects:face detection and deep expression recognition.At the same time,Viola-Jones face detection framework and improved deep expression recognition network based on LeNet-5 are implemented in these two aspects.Finally,the algorithm transplantation and implementation of embedded platform are completed.Embedded realization of deep facial expression recognition is the key content of this paper.In network design,a new activation function and network structure are adopted,and a better facial expression recognition network is trained with the cleaned and expanded FER2013 database.The CMSIS-NN neural network framework is studied on network lightweight and deployment,and two parts of network fixed-point quantization and related calculation of neural network are accelerated.This paper discusses how to deploy a trained facial expression recognition network on embedded devices.Finally,on the basis of theoretical analysis and algorithm research,a low-cost and high-performance embedded machine vision system is designed with STM32H7 processor as the core.Face detection algorithm and deep expression recognition algorithm are implemented in this system,and the real-time and reliability of the system are finally verified.Generally speaking,this paper aims to implement deep facial expression recognition algorithm on embedded edge devices.Through the research and optimization of the algorithm of face detection and deep facial expression recognition,an embedded software and hardware platform for edge computing scenario is built and successfully applied to face expression recognition.On the premise of satisfying real-time and low cost,the facial expression recognition system developed in this paper realizes seven kinds of facial standard emotion recognition,while ensuring accuracy and robustness;the system also has a good human-computer interaction interface and framework system,which provides a set of reference templates for embedded machine vision development;the system ultimately realizes the lightweight of convolutional neural network and the network in the system.Deployment on edge nodes provides a feasible scheme for the application of in-depth learning on edge devices.