AI-Powered Exam Integrity Guardian: Real-Time Classroom Cheating Detection System Using YOLOv8 and Pose Estimation

What Is This Project?

AI-Powered Exam Integrity Guardian is a computer vision system that watches examination halls in real time and flags cheating behavior automatically. It uses two separate YOLOv8 models — one for object detection, one for pose estimation — to catch phone usage, suspicious body posture, and hand-to-hand paper passing between students. Instead of relying on a handful of human invigilators watching dozens of students, the system runs continuous frame-by-frame analysis and logs every incident with a timestamp and confidence score.

Key Features

• YOLOv8 object detection trained specifically to spot mobile phones at varying angles and lighting conditions
• YOLOv8-Pose tracking of 17 body keypoints to detect head turning, unusual bending, and abnormal arm positions
• Hand-proximity calculation between adjacent students to catch document passing in progress
• Seat-position mapping that flags when a student leaves their assigned spot or approaches another desk
• Live webcam monitoring mode with bounding boxes and confidence scores overlaid on the video feed
• Batch video upload with frame-by-frame processing for reviewing recorded exam sessions after the fact
• Configurable email alert system with per-room recipient lists and adjustable confidence thresholds
• Glassmorphism dashboard built with Three.js backgrounds and Chart.js charts for detection trends

How It Works

Video frames from a webcam or uploaded file are passed through the YOLOv8 detection model and the YOLOv8-Pose model in parallel. Detected phones, hands, and pose keypoints are analyzed against threshold rules to classify behavior as normal or suspicious. Flagged incidents are saved to the database with a screenshot, confidence score, and timestamp, and an SMTP email alert is fired to the configured invigilator list. All detections then appear on the Django-powered dashboard for review and reporting.

Tech Stack

• Django 5.0 — handles the backend logic, database models, and admin views with a clean MVC structure
• Ultralytics YOLOv8 — chosen for its speed and accuracy on both object detection and pose estimation without needing two separate frameworks
• OpenCV — manages video capture, frame extraction, and image preprocessing before frames hit the model
• NumPy and Pillow — handle array operations and image manipulation for annotation and cropping
• Chart.js and Three.js — power the dashboard's trend charts and animated glassmorphism background
• SQLite — lightweight storage suited for a project of this scale, easy to swap for PostgreSQL in production
• SMTP email — delivers real-time alerts without needing a third-party notification service

Who Is This For?

Computer science and IT students looking for a final year project with real computer vision depth rather than a toy classifier. Also useful for developers building proctoring tools for online exam platforms, and college administrators exploring automated invigilation for large exam halls.

What's Included

• Complete Django source code with comments
• Pre-trained YOLOv8 and YOLOv8-Pose model weights
• Setup guide and installation documentation
• Sample test videos for demonstration
• Database schema and migration files