CheatGuard

CheatGuard is a lightweight Python-based exam proctoring system designed for offline schools that don’t have access to expensive AI infrastructure or online proctoring tools. It uses OpenCV + MediaPipe to monitor students through a camera and detect suspicious behaviour like:

• Looking away/down repeatedly
• Hands appearing in frame
• Absence of face during the exam
• Detection of prohibited items during the exam

Current Features

• Live video monitoring via webcam
• Face detection to ensure student presence
• Hand detection for questions during the exam
• Automatic photographic and videographic evidence capture when suspicious activity is detected
• Logs alerts to a Flask-SQLAlchemy database with timestamps and photographic evidence
• Saves videographic evidence with timestamps locally
• Built fully using classical CV + MediaPipe (no ML training required)

Tech Stack

• Python 3
• OpenCV
• MediaPipe
• NumPy

How to install

1. Ensure that Python 3.12 is installed. Note: The program will NOT work on a later version of Python.
2. Clone the repository by running the command below in your terminal:

   git clone https://github.com/muditgoel135/CheatGuard.git
   cd CheatGuard
   

3. Create a venv and activate it in your terminal.
4. Download all the dependencies from requirements.txt using the command below.

   pip install -r requirements.txt
   

5. Create a .env file with your secret key with the name SECRET_KEY. Save it as the text below and replace "your_secret_key" with your secret key.

   SECRET_KEY = your_secret_key
   

6. Run the app with the following command:

   python app.py
   

How It Works

1. Webcam feed is captured using OpenCV
2. MediaPipe processes each frame in real-time
3. CheatGuard checks for:
   • Face presence
   • Hand visibility
4. The program draws MediaPipe's given face and hand landmarks
5. If a rule is violated:
   1. An alert is logged.
   2. It is reported to the invigilator.
   3. A frame is saved as evidence

Design Goals

1. Teacher-centric: Final decisions are made by the invigilator
2. Academic-focused: Built for real exam environments
3. Transparent logic: Rule-based detection ensures alerts are interpretable, not black-box
4. Affordable: Works with cheap webcams to support budget-constrained schools
5. Offline-first: Designed for schools without online infrastructure

Current Limitations

• Multi-camera support is not tested
• FPS depends on webcam + system performance
• Lack of simplicity in UI and UX for non-techy invigilators
• Not intended to replace human invigilators (yet 👀)

Future Improvements

• Multi-camera classroom support
• Student ID verification
• Gaze estimation
• GUI for invigilators
• Performance optimisations for low-end systems
• Object detection

Contributing

This project was built fast and intentionally simple. If you have ideas, optimisations, or improvements, PRs are welcome.

License

This project is licensed under the MIT License.