Malpractice Detection System

A real-time exam malpractice detection system using a deep learning model deployed on a Raspberry Pi with a webcam for live monitoring.

How It Works

The system captures live video from a USB webcam, analyzes 16-frame clips using a lightweight MobileNetV2-based model, and flags suspicious behavior when malpractice is detected for more than 8 consecutive seconds. When triggered, it saves a video clip (5 seconds before + 5 seconds after the alert) to the evidence/videos/ folder and plays an audio alert.

Project Structure

├── model.ipynb          # Model training notebook
├── inference.py         # Live detection script (runs on Raspberry Pi)
├── best_model_v4.pth    # Trained model weights
└── requirements.txt     # Python dependencies

Hardware Setup

• Device: Raspberry Pi (any model with USB support)
• Camera: USB Webcam connected to the Pi
• Audio (optional): Place an alert.wav file in the project root for sound alerts

Installation

git clone https://github.com/ThangakumarC/malpractice-detection-system.git
cd malpractice-detection-system
pip install -r requirements.txt

Running Live Detection

python inference.py

Press Q to quit the detection window.

Detection Settings

Parameter	Value	Description
Confidence threshold	0.6	Minimum score to flag as suspicious
Min duration	8 seconds	Time before raising an alert
Cooldown	60 seconds	Gap between consecutive alerts
Pre/Post buffer	5 seconds each	Context saved around alert

Output

Flagged incidents are saved automatically as .avi video files under evidence/videos/ with a timestamp filename (e.g., 20250429_143022.avi).

Dependencies

• PyTorch & TorchVision
• OpenCV
• FastAPI + Uvicorn
• NumPy, Pillow