Automated Cheating Detection in Examination Halls using YOLOv8 and Computer Vision

Ensuring the integrity of examinations remains a significant challenge, particularly in large classrooms where manual supervision is constrained by human fatigue and limited attention. This paper proposes an automated cheating detection framework that leverages computer vision and deep learning to monitor student behavior in real time. A custom dataset was created by recording classroom scenarios with multiple students per frame, extracting frames using OpenCV, and annotating key behaviors with LabelImg. The system employs YOLOv8, a state-of-the-art object detection model, to recognize four primary behaviors: not cheating, turning left, turning right, and using a mobile phone. Experimental results demonstrate that the framework achieves high precision and recall while maintaining real-time processing speed. Limitations include the dataset size, reliance solely on visual cues, and the current inability to detect simultaneous cheating behaviors per student. Future work will expand the dataset, improve detection of overlapping actions, and incorporate multimodal inputs such as audio and pose estimation. The proposed approach provides scalable, unbiased monitoring, reducing reliance on manual invigilation and improving examination fairness.