Drowsiness Detection System Website
A full-stack web application that uses a CNN-based model to detect driver drowsiness in real time through webcam feeds. The Flask backend processes video frames, computes eye-aspect-ratio (EAR) metrics, and triggers visual and audio alerts when fatigue is detected. Features user authentication, database initialization, and a clean dashboard interface.
Tech Stack
The Problem
Drowsy driving is a leading cause of road accidents worldwide. Existing detection systems are expensive hardware solutions. A web-based approach could make drowsiness detection accessible to anyone with a webcam.
Approach
Built a CNN-based Autoencoder for EAR feature extraction, experimented with multiple loss functions, and deployed the model via a Flask web application. The JavaScript frontend captures webcam frames and streams them to the backend for real-time analysis. Added user authentication and a monitoring dashboard.
Key Challenges
- Achieving low-latency inference for real-time video processing
- Balancing model accuracy across diverse lighting conditions
- Building a responsive web interface that works with webcam APIs across browsers
Results
Delivered a working prototype that detects drowsiness in real time with visual overlays and audio alerts, accessible through any modern web browser.