| Popis: |
We propose the use of a particle filter in an automatic user-calibration method for a remote gaze-tracking system. From previous studies, the optical axis of the eye is estimated using two calibrated cameras and two light sources. To realize gaze tracking without user calibration, we must estimate the offsets between the optical and visual axes of the eyes. We propose a method for estimating these offsets automatically. We use a particle filter to calculate the offsets based on the fact that true offsets make no distinction between the points of gaze (POGs) that are calculated from the optical axes of the eyes, irrespective of where the user gazes on the display. We evaluated the proposed method by simulation. Furthermore, we developed a prototype system, which was evaluated experimentally. |
