Eye-Blink Detection Using Facial Landmarks

A real-time algorithm to detect eye blinks in a video sequence from a standard camera is proposed. Recent landmark detectors, trained on in-the-wild datasets exhibit excellent robustness against face resolution, varying illumination and facial expressions. We show that the landmarks are detected precisely enough to reliably estimate the level of the eye openness. The proposed algorithm therefore estimates the facial landmark positions, extracts a single scalar quantity – eye aspect ratio (EAR) – characterizing the eye openness in each frame. Finally, blinks are detected either by an SVM classifier detecting eye blinks as a pattern of EAR values in a short temporal window or by hidden Markov model that estimates the eye states followed by a simple state machine recognizing the blinks according to the eye closure lengths. The proposed algorithm has comparable results with the state-of-the-art methods on three standard datasets.