A Blink Detection Algorithm for Eye-Tracking Applications Based on Facial Geometry

abstract-Blinking is a semi-involuntary physiological process that can provide insight into various cognitive and physiological facets of a human subject, including fatigue or eye conditions. However, in applications such as eye tracking or brain-computer interfaces, blink-related interruptions introduce artifacts and inaccuracies that potentially skew the analysis and interpretation of results. Specifically, blinking poses several challenges to eye-tracking research and applications. Several research groups devised efficient strategies for detecting and minimizing the impact of blinking to improve the accuracy and utility of eye-tracking systems across multiple domains. However, most blink detection algorithms focus on the two-dimensional pixel space of an image. Conversely, this paper illustrates a novel method for blink detection based on the facial geometry extracted from images or videos of the subject. It presents three algorithms for determining the degree of eyelid openness using three-dimensional landmarks of the eye and pupil: eyelid distance, the shoelace theorem, and polygonal methods. Our findings suggest that the proposed blink detection methodology can effectively aid data preprocessing and frame labeling, improving eye-tracking applications. Specifically, our comparative analysis shows that the performances of the three algorithms offer versatility in various contexts where a balance between accuracy and efficiency is crucial.