A Two-Stage, Object-Centric Deep Learning Framework for Robust Exam Cheating Detection
Summary: arXiv:2604.16234v1 Announce Type: cross
Abstract
Academic integrity continues to face the persistent challenge of examination cheating. Traditional invigilation relies on human observation, which is inefficient, costly, and prone to errors at scale. Although some existing AI-powered monitoring systems have been deployed and trusted, many lack transparency or require multi-layered architectures to achieve the desired performance. To overcome these challenges, we propose an improvement over a simple two-stage framework for exam cheating detection that integrates object detection and behavioral analysis using well-known technologies.
Proposed Framework
The proposed framework utilizes two main components:
- Object Detection: The state-of-the-art YOLOv8n model is employed to localize students in exam-room images. This initial stage focuses on accurately identifying the presence of students within the examination environment.
- Behavioral Analysis: Each detected region is cropped and preprocessed, then classified using a fine-tuned RexNet-150 model as either normal or cheating behavior. This second stage ensures that the system can discern between legitimate exam-taking practices and potential misconduct.
Dataset and Performance Metrics
The system is trained on a comprehensive dataset compiled from 10 independent sources, totaling 273,897 samples. The performance metrics achieved by this framework are impressive:
- Accuracy: 0.95
- Recall: 0.94
- Precision: 0.96
- F1-score: 0.95
These results indicate a significant improvement, with a 13% increase over a baseline accuracy of 0.82 in video-based cheating detection. Furthermore, the framework boasts an average inference time of just 13.9 ms per sample, demonstrating its robustness and scalability for deployment in large-scale environments.
Ethical Considerations
In addition to its technical contributions, the AI-assisted monitoring system addresses critical ethical concerns surrounding student privacy. Final outcomes are delivered privately to individual students after the examination, for instance, via personal email. This approach prevents public exposure or shaming, allowing students the opportunity to reflect on their behavior without fear of social stigma.
Future Directions
For further enhancement of the system’s accuracy, future iterations may incorporate additional factors, such as:
- Audio data collected during examinations
- Analysis of consecutive video frames to track behavior over time
These improvements could lead to even more reliable detection of cheating behaviors, solidifying the foundation for developing real-time, scalable, ethical, and open-source solutions in academic settings.