Design And Implementation Of Class Attendance System Based On Face Recognition

Class attendance is a necessary part in the teaching process,the traditional way of attendance will not only increase the teachers' workload,but also occupy a certain amount of time,affect the students in class.At the same time,the development of face recognition technology has great potential in security,identification,automatic monitoring and other fields because of its advantages of direct,friendly and non-aggression.Therefore,this thesis studied face detection,image preprocessing and face recognition algorithm in classroom environment,and designed and implemented the classroom attendance system based on face recognition.The classroom is a place where light is not evenly distributed.At the same time,face image has differentiation in pose,occlusion and so on,so this article chose the multi-task cascaded convolutional networks face detection technology and used it detect and segment face image from the acquised image.To facilitate the later convolution neural network model training in face images of the same position learn a set of reasonable parameters,use image normalization of face image processing,through the five key points(two center of eyes,nose and two corners of the mouth)of image rotation angle,and the original image scaling to 112 × 96 pixel size.After face detection algorithm,the face image is fuzzy,poor quality situation,to facilitate the face classification and recognition in the attendance system,this thesis chooses clarity evaluation algorithm to remove fuzzy,poor quality images.In the feature extraction part,this thesis selects the current extensive convolution neural network in application to construct model.In contrast to the traditional neural network,the convolution neural network has fewer parameters and the training time of network model is shorter,so that it can be used in practical application.Compared with the common softmax loss function,the A-softmax loss function mentioned in this thesis is based on the normalized weight value and the distance of the angle,so that the smallest intra-class distance is greater than the largest inter-class distance.In order to test the convolutional neural networks of different depth in the ability of feature extraction,this thesis set up 10,20 and 36 layer of convolution neural network,on the CASIA-WebFace face library training and testing on LFW face library,with 98.73%,98.73%and 99.13%accuracy respectively.Finally,the class attendance system selects the 20-layer convolution neural network with the highest accuracy.In the end,the hardware structure,system architecture,database and function module of the attendance system are designed according to the requirement of class attendance.On the basis of design,the code writing and user interface the system are completed by using MATLAB and C#hybrid programming.At the end,the function modules of the attendance system are tested on several levels and the demonstration diagram of the system function is given.