Diploma Thesis Review Tereza Soukupová, Eye-Blink Detection Using Facial Landmarks

The thesis presents an algorithm to detect eye blinking of a person in front of a camera. The main idea is to leverage facial landmark detectors that localize characteristic points on the face image. The eye-aspect-ratio (EAR) was introduced as a simple statistic to measure a degree of eye openness. Two methods to detect eye blinks out of the EAR signal over time were proposed: (1) a supervised SVM method, that takes a fixed size temporal neighbourhood and performs a scanning window detection, and (2) an unsupervised HMM-based method that successively adapts to a specific blink pattern of a subject under monitoring. Excellent results achieving the state of the art on standard datasets together with astonishing results in low resolution images were reported.