Modelling of Corneal Surface - A Method for Reconstruction of Corneal Surface in a Wearable Eye Tracking Device

Sammanfattning

We investigate a method for reconstructing the anterior corneal surface using wearable eye tracking devices. The presented method assumes that the four reflection points from two light sources viewed by two cameras lie on a locally spherical surface of a cornea. We then can iteratively estimate the points on the cornea surface by optimizing the local spheres’ center points and radii. Reconstructed surface heights and curvatures are thereafter obtained by polynomial regression. Synthetic data consisting of recordings of images capturing the eyes and corneal reflections from different gaze angles and different corneal shapes are generated for reconstruction and evaluation. The estimated polynomial parameters are compared to the ground truth values of the synthetic data. A spherical cornea, the simplest case used for method verification, can be reconstructed with an order of accuracy of 10−3mm for the cornea radius. We also investigate the impact of different distances between eye and camera setup on the cornea surface reconstruction. For shorter distances, fewer surface points can be computed, and the distance between points of reflection on the surface increases, making the assumption of a local spherical surface less accurate. Therefore, only a lower order of polynomials can be accurately estimated. Contrarily, for longer distances, more surface points can be computed and the assumption of a local spherical surface is more accurate. More accurate and stable results can then be obtained with surface height errors of the order 10−1mm. Given the simple setup using two cameras and ten light sources per eye, the presented method showed great potential for capturing the anterior corneal height and curvature over time. The results should not be generalized to real data without further investigation.