Adaptive prediction of human eye pupil position and effects on wavefront errors

The effects of pupil motion on retinal imaging are studied in this paper. Involuntary eye or head movements are always present in the imaging procedure, decreasing the output quality and preventing a more detailed diagnostics. When the image acquisition is performed using an adaptive optics (AO) system, substantial gain is foreseen if pupil motion is accounted for. This can be achieved using a pupil tracker as the one developed by Imagine Eyes R®, which provides pupil position measurements at a 80Hz sampling rate. In any AO loop, there is inevitably a delay between the wavefront measurement and the correction applied to the deformable mirror, meaning that an optimal compensation requires prediction. We investigate several ways of predicting pupil movement, either by retaining the last value given by the pupil tracker, which is close to the optimal solution in the case of a pure random walk, or by performing position prediction thanks to auto-regressive (AR) models with parameters updated in real time. We show that a small improvement in prediction with respect to predicting with the latest measured value is obtained through adaptive AR modeling. We evaluate the wavefront errors obtained by computing the root mean square of the difference between a wavefront displaced by the assumed true position and the predicted one, as seen by the imaging system. The results confirm that pupil movements have to be compensated in order to minimize wavefront errors.