Abstract
As a key part of the head-mounted eye tracking system, pupil detection not only affects system accuracy, but also system stability. However, the problem of eyelid occlusions arises when eyeball moves. To solve this problem，a Two-Level pupil detection method was proposed. The first level utilizes the improved starburst method to extract pupil edge points and then check whether the pupil is shaded by eyelid. If its not the case, pupil detection ends, otherwise the second level detection is conducted. In the second level detection, improved ellipse template matching method is applied. Unlike other ellipse matching methods, this method is based on a 3D eyeball model. With eyeball parameters, this method estimates ellipse minor-major axis rate and angle of rotation in advance, and then decreases the spatial complexity from five dimensions to three, thus improving matching efficiency. With our method, the pupil can be easily rapidly detected under eyelid occlusions.
Abstract
As a key part of the head-mounted eye tracking system, pupil detection not only affects system accuracy, but also system stability. However, the problem of eyelid occlusions arises when eyeball moves. To solve this problem，a Two-Level pupil detection method was proposed. The first level utilizes the improved starburst method to extract pupil edge points and then check whether the pupil is shaded by eyelid. If its not the case, pupil detection ends, otherwise the second level detection is conducted. In the second level detection, improved ellipse template matching method is applied. Unlike other ellipse matching methods, this method is based on a 3D eyeball model. With eyeball parameters, this method estimates ellipse minor-major axis rate and angle of rotation in advance, and then decreases the spatial complexity from five dimensions to three, thus improving matching efficiency. With our method, the pupil can be easily rapidly detected under eyelid occlusions.
Open Access
JUSTC
DownLoad: