Implementation Of Face Recognition System Based On Resnet

With the rapid development of intelligence,people have higher requirements for the security and convenience of various occasions,so face recognition technology has a wider range of application scenarios.With the continuous improvement of chip computing capabilities,especially under the trend of mobile internet,various embedded intelligent devices have become more deeply embedded in people's lives.This brings great opportunities to the development of embedded face recognition systems.In this context,this paper designs and implements a face recognition system based on ResNet.This paper completes the design of multiple functions including face collection,establishment of face data set,and network model training.Among them,Haar feature plus Adaboost classifier is used as the main algorithm of face detection.Faces were collected through video streaming,and a small data set of 10 people with different poses and expressions was established.In the face matching model,a ResNet neural network with the Bottleneck jump connection structure as the core unit is designed,and the Batch Norm layer and the Dropout layer are used together to prevent the model from overfitting.An ARM embedded platform based on I.MX6 U is designed.From the perspective of system functions and actual requirements,the software and hardware of the system are designed in detail.The hardware design consists of designing the system's overall hardware structure.This system uses I.MX6 U as the central processing unit chip,and uses high-precision CMOS OV5640 camera to collect images.In terms of software design,the Open CV library and the Tensor Flow architecture were installed and transplanted on the host computer and the host computer,respectively,and the humancomputer interactive operation interface was compiled using the Qt library.Finally,a host computer simulation training was formed,and the host computer runs reasonably and efficiently System development environment.Based on the self-built face data set,the whole system was debugged and the results analyzed.The experiment results show that The recognition speed and accuracy of the system have achieved good results,fulfills the design requirements,and has certain use value.