Design And Implement Of Face Recognition System Based On Multi-angle Video

With the rapid development of science and technology,information security has been challenged by new technologies.People are turning their attention to biological authentication technology with higher security,such as fingerprint recognition,iris recognition,face recognition and other technologies.The face recognition has the advantages of non-contact,non-mandatory,and higher accuracy,which has been applied to the area of payment system,security system,access control system and other application systems.At present,in the security system of station and access control system of community,the users need to stop and watch the camera.In order to broaden the application of face recognition technology,we design a face recognition system based on multi-angle video.The system can monitor from wider range and multi-angle to collect more face images,the users do not need to stop and watch the camera.The users only need walk freely within a given space to be recognized.This paper mainly studies the face recognition method based on multi-angle video in the relatively closed indoor scene,and implements the system.The main work of this article is as follows:(1)We design the architecture and solution of face recognition system based on multi-angle video.we build a face recognition channel which is composed by multiple cameras.We determine the location of the camera by a lot of experiments,and implement a face recognition system.(2)We design a face detection method which is combined with AdaBoost and CNN to solve the problem of high error detection rate for AdaBoost face detection method based on LBP features.The AdaBoost algorithm is used to exclude most of the non-face windows.Then,a deep learning method is used to make two judgements for a small number of face candidate windows,so as to preserve the face window and exclude the non-face windows,and take account of the balance between speed and accuracy.(3)An improved lightweight convolution neural network model is designed for extracting the face feature to solve the problem of slow computation on CPU based on deep learning model.We replace the connected layer by convolution layer to implement a lightweight fully convolution neural network.By this way,we reduce the parameters of the model and shorten the time of extracting the face feature.