Purchase this article with an account.
Alberto De Castro, Xiaofeng Qi, Lucie Sawides, Stephen A Burns; Adaptive Optics Control Algorithm to Detect the Pupil and its Boundary in real time Using Shack Hartmann Images. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):5981.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To automate pupil detection and tracking in an Adaptive Optics (AO) system using Shack Hartmann (SH) data. To automatically choose pupil position and shape so as to maintain high quality AO control.
A spot quality metric was defined for each SH lenslet. Spots with low quality metrics occur either because the eye’s pupil is not in that location or because there is an obscuration or a corneal or lenticular problem. For this study we defined the metric for each SH spot as the ratio between the intensity of a region around the spot peak equal to 30% of the CCD space, and the total intensity. We then automatically detect the boundary of the pupil. Two approaches to minimize the influence of the defective spots in the final shape of the Deformable Mirror were studied. 1) The wave front at the locations of missing spots was assumed to require no correction. 2) The missing spots were eliminated from the influence function matrix by removing the corresponding rows.<br /> The exit pupil of an AOSLO was magnified to subtend 10.5 mm at the eye. Two subjects were imaged with dilated pupil sizes of 7 and 6 mm diameter. Image quality based on either mean intensity or spatial frequency content of the images was calculated. Results obtained when sampling the whole exit pupil using the new algorithms were compared to results from our standard approach, which requires sampling only the subject’s pupil and keeping it centered in the system.
The quality of the images captured when the defective spots are removed in real time from the influence matrix was 99% of the intensity using the standard algorithm. The area under the FFT of images was 97%. If “missing” spots were not removed but set to zero deviation (the ideal spot positon) images were darker (89%) and had lower frequency content (92%). As reported previously the images improved when the spots on the boundary of the pupil were marked as defective in either algorithm (intensity 89 to 99% and 99 to 102%, frequency content 92 to 97% and 97 to 99% respectively).
It is possible to relax the constraints on head positioning in Adaptive Optics imaging by using an adaptive algorithm. There is little impact on image quality and the algorithm chosen can operate in real time (30 Hz AO loop).
This PDF is available to Subscribers Only