Face detection based on image stitching for class attendance checking

Abstract

Traditional attendance checking relies on teachers to name students in the class, which has many shortcomings, including time wastage and fake sign-ins. We use an online system to allow the students to sign in and take advantage of the face detection technology to help teachers check attendance in their classes. Because large classrooms require teachers to take multiple photos to include all students, picture stitching is required before face detection can be performed. We need to solve the problems of image registration, image fusion, adaptation of shooting angles, the proportion of overlapping parts of images, and the problem that YOLOv3 cannot detect small faces. Our research focuses on achieving a good result in stitching and improving the accuracy of face detection. The proposed system has tried to give teachers some good advice on photo shooting angles and overlapping proportions of multiple photos. Also, we make some improvements in feature point extraction and small faces detection algorithm etc. The experimental results show that the system can identify nearly 100% faces in the photos.