ProctorAI is an intelligent, multimodal AI system designed to ensure the integrity of remote assessments through real-time behavioral understanding. Developed initially as a Master's research project at Kwara State University, this system moves beyond simple binary "cheating" flags by leveraging environmental perception to detect disruptive sounds and unauthorized visual infractions. https://www.bravía.com/academic-writings/dissertation/proctoring-dissertation.pdf.
The core philosophy of this project is advancing Trustworthy and Explainable AI in education, reducing false positives, and providing context-aware anomaly detection.
- Real-Time Video Anomaly Detection: Utilizes computer vision to monitor head pose, gaze diversion, and unauthorized persons in the frame.
- Audio Signal Processing: Detects anomalous background noise and secondary speech via WebRTC data streams.
- Context-Aware Logging: Records system-level events alongside audio-visual data for comprehensive behavioral analysis.
- Explainable Outputs: Rather than opaque decision-making, the system aims to provide confidence scores and contextual justifications for flagged events.
ProctorAI operates on a decoupled architecture designed for scalability and real-time processing.
- Backend / AI Microservice: Python, Flask, OpenCV
- Frontend: HTML/JS (Currently migrating to a Next.js / React architecture for enhanced state management and scalability)
- Data Streaming: WebRTC for low-latency audio/video transmission
- Infrastructure: Deployed on Oracle Cloud Infrastructure (OCI) Free Tier.
- Note on v2 Migration: We are actively integrating Celery and Redis to handle heavy, asynchronous AI inference tasks in the background, preventing frontend blocking during intensive frame analysis.
The methodologies and foundational research for this repository have been published in peer-reviewed journals:
- An Enhanced Web-Based Examination System using Automated Proctoring and Background Activity Detection. (JIRBDAI) https://www.bravía.com/academic-writings/publication/1-toyyib-et-al.pdf.
- Implementing an AI-Driven Proctoring System: Real-Time Detection of Disruptive Sounds and Unauthorized Visual Infractions. (IJERD) https://www.bravía.com/academic-writings/publication/4-olanrewaju-et-al.pdf.
This repository serves as the foundation for upcoming doctoral research focused on Multimodal Fusion and Trustworthy AI. Planned architecture updates include:
- Implementing temporal sequence modeling (LSTM/Transformers) to analyze behavior over a 10-second sliding window rather than isolated frames.
- Integrating a dedicated explainability layer (e.g., SHAP) to generate human-readable justifications for automated flags.
- Full migration to Next.js with a Node.js signaling server.
(Provide brief instructions here on how to run your code locally. Example below:)
# Clone the repository
git clone https://github.com/braviadev/proctorai.git
# Navigate into the directory
cd proctorai
# Install dependencies
pip install -r requirements.txt
# Run the Flask development server
python run.py