This study aims to develop of an intelligent system for detecting cheating in online examinations through real-time video stream analysis using deep learning and computer vision techniques. Three independent classification models were developed, each differing in input types and architectural structures, and subsequently converted into the ONNX (Open Neural Network Exchange) format to ensure cross-platform compatibility. To achieve real-time responsiveness, the system was deployed through a multi-threaded interactive web-based interface. Additional modules were integrated to enhance functionality. These include suspicious objects detection using YOLOv9, retrained on a custom examination dataset to improve detection accuracy and better identify cheating behaviours, as well as mouth movement tracking based on the Mouth Aspect Ratio (MAR) indicator to determine whether students were speaking during the exam. The experimental results demonstrated that the hybrid model outperformed the other models, achieving a classification accuracy of 98.8% with a complete absence of false negatives. These findings reinforce the reliability of the system in sensitive examination settings and support its adoption in digital educational platforms.
Homs University Journal.
2026.
Online Exams Cheating Detection Based on Video Analysis Using Deep Learning