Abstract
Pupillometry (or measurement of pupil size) is commonly used as an index of cognitive load and arousal. Pupil size data is recorded using eye-tracking devices that provide an output containing pupil size at various points in time. During blinks, the eye-tracking device loses track of the pupil and this results in missing values in the output file. The missing-samples time window is preceded and followed by a sharp change in the recorded pupil size, due to the opening and closing of the eyelids. This eyelid signal can create artificial effects if it is not removed from the data. Thus, accurate detection of the onset and the offset of blinks is necessary for pupil size analysis. While there are several approaches to detecting and removing blinks from the data, most of these approaches do not remove the eyelid signal or they result in a relatively large data loss. The current work suggests a novel blink detection algorithm based on the fluctuations that characterize pupil data. These fluctuations ("noise") result from a measurement error produced by the eye-tracker device. Our algorithm finds the onset and offset of the blinks based on this fluctuation pattern and its distinctiveness from the eyelid signal. By comparing our algorithm to three other common blink detection methods and to two independent human raters, we demonstrate the effectiveness of our algorithm in detecting blink onset and offset.
Meeting abstract presented at VSS 2018