An automated proctoring assistant in online exams using computer vision

Abstract

Cheating or attempting to cheat in education has had the opportunity to become numerous and complex since the outbreak of the COVID-19 pandemic, with teaching and testing conducted online. The learners have easier access to prohibited materials, and it is easier to avoid contact with a human proctor. Such problems raise the need for an automated intelligent proctoring system to help the teacher supervise students. Therefore, this work proposes a system that can automatically examine students' behaviors through two main cameras. The first camera takes images of a student's frontal face and uses them as input for the facial landmark model, detecting anomalies in a student’s face movement. The second camera captures the student’s whole body and the surrounding environment, and by using a trained pose recognition model, the system can efficiently classify student actions as suspicious or not. In addition, an object recognition model is also applied simultaneously to detect people and objects appearing through the second camera. Results of this research show good remarks and can be applied in schools, universities experimentally in the future.