Design And Implementation Of Class Attendance System Based On Face Recognition With Real-Time Video

Class attendance checking is one of the effective methods of classroom management,it can effectively supervise students to attend classes on time and to ensure the quality of classroom teaching.At present,students’ attendance is accomplished through roll calling and randomly spot checks by the teaching staff.This traditional manual roll calling method not only consumes a lot of manpower and time,but also cannot detect lateness,early left,substitution and absenteeism in real time,and cannot fully supervise student attendance.In recent years,deep learning has made great progress in static face image recognition,which provides a new idea for the realization of automatic classroom attendance system.However,due to the limitation of classroom space,most surveillance cameras are installed in front(or rear)of the classroom.The relative position of the students from the camera is quite different,and the pixels of face images picked from those who are far away from the cameras are lower.The face features are not obvious after normalized clipping,which will seriously affect the students’ correct recognition rate.In classroom environment,the phenomenon of human body moving and posture deviation will lead to problem of motion blur in some face region images detected in the sampling frame.The face region image with motion blur will also bring the defect that face feature is not obvious.The above phenomena will seriously affect the performance of the recognition algorithm,which leads to the deterioration of the performance of the whole class roll calling system.Therefore,in classroom environment,how to ensure that the size of face image meets the requirements of recognition by means of control,how to select high quality face region image in video stream and how to obtain the final recognition result by using the single recognition result statistic in the sampled frame effectively are three main problems studied in this paper.The specific solutions are as follows:(1)In this system combined with classroom scene segmentation and camera configuration proposed in this paper,the absolute position information of human face is obtained by multi-task convolutional neural networks(MTCNN),and the control method of cloud head based on face target search is completed,which can effectively guarantee the image size of face region.(2)The image quality assessment method is introduced into the traditional face recognition system in this paper,through this method,the images with obvious facial features in the sampled frames are automatically selected to ensure the effectiveness and robustness of the face recognition system in the classroom video stream.(3)After ensuring face image size and quality,the face features are extracted by the improved FaceNet structure,and the single recognition results are obtained by the SVC(Support Vector Classifier).Combined with multi-efficient single recognition results,a multi-effective frame based probability voting method is also proposed.The final recognition results obtained by this method are relatively stable.