An Intelligent Real-Time System for Exam Cheating Detection using Computer Vision and Deep Neural Networks

The rapid digital transformation in education has introduced new challenges to academic integrity during examinations. Traditional monitoring that relies solely on human supervision is no longer sufficient due to the large number of students and increasingly sophisticated cheating methods. This study aims to develop an integrated real-time cheating detection system based on computer vision and deep learning. The system combines object detection (YOLOv11), pose estimation (Media Pipe), head orientation analysis using the Perspective-n-Point algorithm, and (Ghostface Net) for facial recognition.

The system processes live video streams to identify suspicious behaviors such as looking away, hand proximity to the face, and interaction with unauthorized objects like mobile phones and books. A multi-component suspicion scoring mechanism assigns relative weights: head orientation (75%), hand proximity to the face (15%), and hand-object interaction (10%), producing a unified suspicion score. The system sustains real-time operation (16.98 FPS), ensuring smooth display and immediate responsiveness through a Flask-based interface, while reducing false alarms via temporal smoothing with Exponential Moving Average (EMA) and a deduplication mechanism.

Furthermore, the system adheres to privacy standards by storing only suspicious frames and automatically deleting data to safeguard student information. This work represents a practical and academic contribution to strengthening academic integrity and reducing human errors in exam monitoring.