Abnormal Behavior Detection in Electronic-Exam Videos Using BeatGAN

Abnormal Behavior Detection is a core ability to recognize cheating in electronic exams (e-exams), especially in the scenes where a fraudulent candidate hides unauthorized resources from the view of the proctor. This work utilizes the BeatGAN algorithm to detect behavioral abnormalities in the skeleton-based motion information of candidates during remote e-exams. We examine the algorithm on the previously published dataset of skeleton features extracted from e-exam videos. The results show that BeatGAN outperforms previously investigated algorithms in sequence anomaly detection using the raw joint-coordinate, implicit bone-length-and-angle and explicit bone-length feature types. But, in event anomaly detection, BeatGAN does not improve the state-of-the-art results on the dataset.