AI & Automation
Blink to Text
Real-time blink-detection application that converts eye blinks into text so people with Locked-in Syndrome can communicate independently on a low-end computer and webcam.
Role: Co-author (B.Tech minor project team, Maharaja Surajmal Institute of Technology)
Situation
People with Locked-in Syndrome remain cognitively aware but lose nearly all voluntary movement except eye motion and blinking. Existing eye-controlled tools often depend on expensive hardware.
Decision question
Real-time blink-detection application that converts eye blinks into text so people with Locked-in Syndrome can communicate independently on a low-end computer and webcam.
Data and constraints
See method and results below for sample, coverage, and documented limits.
Method
Compared Viola-Jones and HOG + SVM face detection, then used dlib 68-point facial landmarks and eye-aspect ratio to detect blinks. Built an Electron desktop GUI (HTML, CSS, JavaScript) with a Python OpenCV/dlib backend that shares blink state over a local TCP socket so the interface can be started, paused, and operated with blinks alone.
Tools
- OpenCV
- dlib
- HOG + SVM
- Haar Cascade
- Eye-aspect ratio
- Electron
- Python
- JavaScript
Deliverables
- Research paper on real-time blink detection and text display
- B.Tech minor project report (December 2018)
Outcome / decision value
- Selected HOG + SVM over Viola-Jones because it is more robust to non-frontal pose and illumination for users who cannot stay facing a camera.
- Designed the system so it can be started, paused, and operated entirely with blinks, reducing dependence on caretakers for communication.
- Targeted low-end computers and a basic webcam, including use on devices such as a Raspberry Pi, rather than specialized eye-tracking hardware.