Class Attendance Checking System based on Deep Learning and Global Optimization

This paper proposes a class attendance checking system based on deep learning approach. Because classrooms are non-constrained environments and each student often has only one sample photo in the learning management system, traditional methods are difficult to achieve good results. Considering the deep learning methods' excellent performance in complex scenes, we build a class attendance checking system using deep learning methods. First, the Faster-RCNN model is used for face detection. Then, we track the detected faces and extract the features of the face images through a CNN model. After calculating the distance between the detected face features and the sample face features, a global optimization method is used to accomplish the recognition of all face tracks at the same time. This paper begins with an introduction to the application background, and reviews the related works in the field of face-recognition-based attendance checking in the classroom. Then, this paper introduces our system architecture and the methodology used in each section. Finally, we test the system on real class surveillance videos. The results demonstrate the validity of our system architecture and the methods we used.