AI-Powered Cheating Detection System for Offline Exams
EyeSpy is an AI-powered object detection system designed to detect cheating in offline classroom exams using real-time video surveillance.
Unlike online proctoring tools, EyeSpy is focused on physical classrooms where invigilation is still manual and error-prone.
- Framework: YOLOv8 + PyTorch
- Dataset: 880+ labeled images (cheating & non-cheating scenarios)
- Accuracy: Achieved 92% precision and 85% recall
- Deployment: Supports ONNX export for cross-platform compatibility
- 📷 Real-time Cheating Detection using webcams
- 🏷 Binary Classification: Cheating vs Non-cheating
- 🔄 Data Augmentation for robust performance
- 🚀 Lightweight YOLOv8s model for fast inference
- 📊 Performance Metrics: Precision, Recall, F1-score, Confusion Matrix
- Programming: Python 3.8+
- Frameworks: PyTorch, Ultralytics YOLOv8
- Data Processing: Albumentations, NumPy, OpenCV
- Training: Kaggle GPU (Tesla T4)
- Precision: 0.92
- Recall: 0.85
- F1-Score: 0.89
- Model performs real-time inference with high accuracy in classroom scenarios.
- GUI with real-time alerts
- Multi-class cheating classification
- Mobile/Edge deployment
- Improved robustness to lighting & camera angles
You can read our detailed project report here:
📄 EyeSpy MidTerm Report (PDF)
- Hridayanshu Raj Acharya
- Bhishm Pd. Bhandari
- Ashraya Kadel
- Parth Pandit
- Sachin Shrestha
This project is for academic use only.