System Calibration of Visible-Spectrum Gaze Tracker With Linear Models

Calibration for infrared-based gaze trackers, which maps the estimated iris center locations in the image to monitor coordinates, is generally achieved with second-order polynomials. However, calibration performance tends to be poor for visible-spectrum gaze trackers (VSGT) due to the absence of corneal reflections from infrared light sources. This paper introduces a novel calibration method for VSGT that utilizes linear models based on the polar coordinate system. Experimental results demonstrate that false gaze data can be automatically filtered out through linear regression, significantly enhancing system accuracy.