Automated Academic Cheating Detection Using Video Surveillance and Circumstantial Action Recognition

Abstract

Due to the aftermath of COVID-19 pandemic, educational institutions worldwide have transitioned from traditional offline teaching to online methods. Unfortunately, some students resort to unfair practices during online exams, compromising the integrity of the assessment process. In this paper, a cheating detection framework has been proposed which is specifically designed for online exams. The framework relies on eight types of Circumstantial Actions extracted from video sequences. Initially, a dataset consisting of image frames are generated from the captured video sequences. Thereafter, various feature extraction methods are employed on those images to generate the feature vectors. Finally, combination of K-Means Clustering in association with Support Vector Machines (SVM) is used to classify the data for identification of cheating in exam. Analysis of the experimental results indicate that the proposed method achieves a notably high level of accuracy in detecting cheating during exams.