Using Morphological Filters for Pupil Detection in Infrared Videos

Abstract

Abstract : Patterns of eye movements may be indicative of a subject's cognitive state, including the likelihood that a subject is lying or telling the truth. Medical systems exist for tracking eye movements, but their use requires artificially constraining head movements in a manner that is incompatible with an interview scenario. The eye movement analysis contained in this report was conducted retroactively, using data collected elsewhere using a simple experimental setup. The subjects were photographed using a single camera equipped with a single IR (infrared) illuminator. A three-stage process of morphological filtering was used to detect the subjects' pupils, with manual parameterization of the filter coefficients. False detections due to background clutter from eyeglasses and jewelry were significantly reduced, but not entirely eliminated. Use of an elliptical filter, rather than a circular filter, improved the outcome for the final stage of pupil detection. Although improved methods exist for reducing false detections, the data collection protocols are far more complex than the single-camera/single-illuminator protocol used here.