Research On Key Technologies Of Automatic Attendance System Based On Face Recognition

In this thesis,an automatic attendance system based on face recognition is implemented for the classroom scene.Based on the surveillance video stream of the classroom,system firstly detect faces in the classroom and get a face image set,secondly the face filtering method is used to eliminate faces of low quality or in the position that has been successfully identified,Thirdly system perform face super-resolution and face alignment on face images,finally face images are sent to the face recognition model for face comparison to complete the attendance.The main research work of this thesis is as follows:(1)Proposed an improved Retain Net face detection model based on classroom scenes.There are three main improvements: Firstly,use DIo U Loss as the loss function of face frame regression to improve the intersection ratio between the predicted frame and the real face frame;Secondly,add a CBAM module based on the hybrid attention mechanism to the network to better extract facial features;thirdly,propose an anchor frame setting strategy based on the ROIIC(Region of Interest in classroom)to add additional anchor frames to assist detection,Experiments show that the improvement proposed in this paper effectively improves the accuracy of classroom face detection.(2)A image filtering method for the set of faces to be recognized based on face position and face quality is proposed,which removes face images whose recognized position and face quality are not sufficient to support subsequent face recognition,and improves system efficiency;(3)To solve the difficult problem of low solution face recognition,a multi-task model of joint training of face super-resolution and face key point detection is proposed,which effectively improves the accuracy of face recognition.(4)Based on the above research,an automatic attendance system based on face recognition is designed and implemented,which can realize automatic attendance based on the existing cameras in the classroom,or help students find free self-study classrooms.The experimental results show that the solution in this paper is effective.It can meet the attendance needs of most classroom scenes.