Face Recognition Research And Design Based On Deep Learning In Monitoring Scene

With the rapid development of computer vision technology,face recognition technology has come into people's life.One of the most important applications of face recognition is real-time face recognition in the monitoring scene,which is the most difficult.Real monitoring scene has a lot of challenge of uncertain factors.,which contained that collected a large amount of data,how to face recognition based on a small amount of data,faces tend not to take the initiative to consciously to the camera,video camera under the face have different attitude,different light intensity,low resolution and size and still keep out.Specifically,the research content of this paper can be divided into the following four parts:1.Research on the loss function of face recognition,In the real monitoring scene,because of the complex environment conditions,the traditional softmax loss function can not be used to correct classification.We proposed the i-center loss to reduce the distance between classes,increase the distance between classes.Based on MNIST,CIFAR10 and LFW,we did some comparative experiments,besides,we finished the visual analysis.2.Research on face feature extraction,we puts forward a very innovative idea and proposes a new face recognition model Mir-Net.We believe that different individuals have commonalities,and the commonalities between faces can be extracted.Then,the remaining features of different faces can be used for classification,which will improve the accuracy of face recognition.3.Research on face detection,In this paper,features of different levels are integrated to make the network acquire features of high resolution and strong semantics so that we can deal with real and complex monitoring scenes to improve the accuracy of face detection.4.Implementation of face recognition system,The real-time video face recognition system can be built for online and of:fline face recognition.The system can receive the camera data to finish the face recognition.In this paper,a new face recognition network model,Mir-Net,is proposed to perform face recognition on a small amount of data by means of transfer learning and the loss function of face recognition is improved to further improve the accuracy of face recognition.Finally,on the basis of theory and experiment,a real-time video face recognition system is designed and implemented in this paper.