Toward Flexible Calibration of Head-mounted Gaze Trackers with Parallax Error Compensation

Although the mobile head-mounted gaze tracker (HMGT) has gained its great success in human-machine interactions, the real implementation of HMGT still poses several significant challenges. The parallax error and the tedious calibration procedure, as two of these challenges, will be addressed in our proposed two-step calibration method. In the first step, instead of fixating at several pre-defined calibration points successively, the user is only required to change his or her head pose while gazing at one calibration marker with allowance of short-period distractions. Users can continuously change head poses or keep each of their head poses for 1 similar to 2 seconds before changing to the next head pose. Then the traditional Gaussian process (GP) regression considering the dominant linear trend and a sparse GP regression using pseudo-inputs are applied for above two kinds of head motions to register input space of glint-pupil vectors with output space of image gaze points, respectively. Besides, the heat map, the correlation test and blob analysis are made use of to detect users' distractions and eye blinks during the calibration phase. The second step involves recovering the epipolar geometry set up by HMGT and eyeballs. To this end, a nonlinear optimization problem is formulated to search for the geometry parameter set which could well explain the detected parallax errors resulted from varying the depth of the calibration point. Then the real image gaze point can be straightforwardly estimated as the intersection of two epipolar lines calculated from binocular data. The simulation and experimental results validate the effectiveness of our proposed calibration method.