Pupil Detection Using Stereo-Matching Method and a Constant Interpupillary Distance Condition for a Solution of Glasses Reflection Problem in the Video-Based Gaze Detection System

In the pupil-corneal reflection detection-based eye-gaze detection method, glasses reflection of near-infrared LED light sources for producing the corneal reflection is misdetected as the pupil when a user wears eyeglasses. To improve the robustness of the pupil detection, we propose novel pupil searching and tracking methods in the gaze detection system using two stereo-calibrated cameras. The pupil searching method first chooses the true pupils from all stereo-matched pupil candidates using the suitable depth range condition, and second chooses the true pair of the right and left pupils under the constraint of the suitable 3-D interpupillary distance. Even if one pupil is not detected in the image of either camera owing to the glasses reflections, the pupil tracking method estimates the 3-D coordinates of the undetected pupil by using the constant interpupillary distance and the temporal continuity of the 3-D coordinates of the moving pupil. The experimental results show that the accuracy of pupil searching and tracking was better than that of the conventional one-camera method.