A Novel approach to a wearable eye tracker using region-based gaze estimation

Eye tracking studies are useful to understand human behavior and reactions to visual stimuli. To conduct experiments in natural environments it is common to use mobile or wearable eye trackers. To ensure these systems do not interfere with the natural behavior of the subject during the experiment, they should be comfortable and be able to collect information about the subject\u27s point of gaze for long periods of time. Most existing mobile eye trackers are costly and complex. Furthermore they partially obstruct the visual field of the subject by placing the eye camera directly in front of the eye. These systems are not suitable for natural outdoor environments due to external ambient light interfering with the infrared illumination used to facilitate gaze estimation. To address these limitations a new eye tracking system was developed and analyzed. The new system was designed to be light and unobtrusive. It has two high definition cameras mounted onto headgear worn by the subject and two mirrors placed outside the visual field of the subject to capture eye images. Based on the angular perspective of the eye, a novel gaze estimation algorithm was designed and optimized to estimate the gaze of the subject in one of nine possible directions. Several methods were developed to compromise between shape-based models and appearance-based models. The eye model and features were chosen based on the correlation with the different gaze directions. The performance of this eye tracking system was then experimentally evaluated based on the accuracy of gaze estimation and the weight of the system