Eye movement tracking for diagnostic systems

This dissertation presents an image based eye tracking system for human disease diagnostic/screening systems. Current eye tracking systems either do not satisfy the speed or cost requirement, or are not robust enough for clinical use. This dissertation focuses on developing an accurate, robust, and fast online eye tracking system.;The dissertation presents an accurate, robust, and high-speed horizontal and vertical eye tracking system on a general computer system. A high frame rate of 150Hz for binocular systems is achieved by applying multi-resolution image processing techniques. A novel symmetric mass center algorithm is developed to handle occlusions. A torsional eye tracking system robust to interference is presented as well. Multiple mechanisms in the calibration and tracking phase are designed to ensure the robustness of the system. Simulation results show significant improvement in accuracy and robustness.;Diabetic retinopathy (DR) is one of the potential complications of diabetes that may cause blindness. This dissertation presents an accurate and online pupil size monitoring system for the screening of DR based on the difference of pupil response to peripheral and central light stimulus. Test results on pupil area measurement show that the designed system provides a more accurate measurement of the pupil size under interference. The dissertation also presents a proof of concept test on human subjects to verify the screening method using the pupil monitoring system developed; the results show that there is a clear correlation between the normalized value of constriction velocity, amplitude and the severity of diabetic retinopathy, which was a hypothesis before.;Discovering the 3D eye position from 2D images requires the recovering of image projection parameters through calibration. In contrast to existing work, we use the more accurate two-radius rotation model for eye movement and the perspective projection model for the imaging process. The parameters of the elliptical pupil boundary from multiple views of the eye are used as the constraints to estimate the projection parameters. Simulation results show remarkable improvements in measurement accuracy. The simulation system is designed to establish ground truth for the eye position in the tests; it synthesizes realistic eye images by simulating the imaging process.