Anomaly Detection System for Possible Cheating During Online Examination

Online exams are vulnerable to cheating due to the absence of direct supervision. This study uses webcams to present a system for detecting smartphone use and off-screen gaze during online exams. Leveraging YOLOv8 for smartphone and face detection and MediaPipe for head pose estimation, the system analyzes live web cam feeds in real-time anomaly detection. We trained the system using a merged dataset derived from public Roboflow datasets. The system achieved high accuracy, with a mean average precision (mAP@O.5) of 97.9% for smartphones and 99.5% for faces. Head pose estimation demonstrated low mean absolute errors (MAE). This integrated approach enhances academic integrity in online assessments, providing a real-time monitoring tool for educators.