Predictive Compensation And Image Registration In Adaptive Optics Systems Based On Motion Estimation

Adaptive optics(AO)system uses wavefront sensors to measure wavefront distortion,and calculates control signals by the controller,then the wavefront corrector generates corresponding shapes to compensate distorted wavefront according to the control signals.Adaptive optics system has been widely used in astronomical imaging and ophthalmic imaging.In the astronomical imaging AO system,under the atmospheric frozen flow turbulence assumption,atmospheric frozen flow turbulence is not static but is driven by the transverse wind.The time delay in the system results in a mismatch between the correction shape generated by the wavefront corrector and the actual distorted wavefront.This kind of correction error is time delay error.In the retinal imaging AO system,the eyeball cannot be completely fixed during the imaging.The fixational eye motion during the imaging causes the relative deformation between the imaging results and the eye motion artifacts in the imaging results.This thesis focuses on the predictive control and image registration in AO systems using motion estimation.To compensate the time delay error in the astronomical imaging AO system,the predictive control with the atmospheric transverse wind estimator is proposed.To compensate eye motion artifacts in adaptive optics optical coherence tomography(AO-OCT)system for retinal imaging,an intra-volume eye motion artifacts correction method is proposed.The main contents of this thesis are consisted by five parts:In the first part,Shack-Hartmann Wavefront Sensor(SHWFS)is used to measure the transverse wind using the reconstructed wavefront from the SHWFS slope measurements.The relationship between the estimation error and the reconstruction order is analyzed,and the optimal reconstruction order range is obtained.Two transverse wind estimation algorithms are proposed: block match method and optical flow method.Two methods are compared using estimation bias and variance.For both methods,the percentage estimation errors of transverse wind velocity are kept within 30%,and the estimation error of transverse wind direction is kept within 6 degrees.The estimation variances of the two methods increase with the variance of the slope measurement noise.The variance of these two methods are similar.The optical flow method is more preferable since it has smaller estimation bias than the block match method.A rotating phase screen atmospheric turbulence simulation device is used to verify the transverse wind estimation methods using reconstructed wavefront from the SHWFS slope measurements.The results show that the percentage estimation errors for two atmospheric transverse wind speeds are both within 10%.In the second part,a transverse wind estimation algorithm using SHWFS slope measurements is proposed.The Cramer-Rao lower bound(CRLB)is used to analyze the fundamental performance limit of transverse wind estimator from SHWFS slope measurements.By analyzing the Fisher information of the transverse wind estimation,it is found that when atmospheric coherence length is not less than 3/4 size of SHWFS subapertures.The unbiased CRLB is negative proportional to atmospheric coherence length.Unbiased CRLB is proportional to the standard deviation of the slope measurement noise.Then the bias of the transverse wind estimator based on the gradient is analyzed.The deterministic bias is caused by the neglection of the high order terms in linearization and the gradient approximation error.Finally,a biased CRLB based on gradient-based transverse wind estimator is obtained.This performance bound can predict the estimator performance more accurately.A rotating phase screen atmospheric turbulence simulation device is used to verify the transverse wind estimation methods using the SHWFS slope measurements.The results show that the percentage estimation errors for two atmospheric transverse wind speeds are both within 15%.In the third part,the predictive control with transverse wind estimator is proposed to compensate the time delay error in AO system.The method includes two parts: atmospheric transverse wind estimation and prediction correction.Atmospheric transverse wind estimation using the SHWFS slope measurements.The wavefront prediction is realized in the Fourier domain using the estimated atmospheric turbulence transverse wind parameters.The prediction error of SHWFS subapertures in the aperture edge is analyzed and compensated.With the known transverse wind parameters,the proposed prediction method can almost completely compensate the time delay error.Then the robustness of the predictive correction method with respect to the transverse wind estimation error is analyzed.The theoretical tolerance range of the estimation error for transverse wind velocity and direction is obtained.When transverse wind estimation errors of velocity and direction are both existed,predictive correction can improve the AO system's correction performance in a large transverse wind estimation error range.The correction performance of the predictive control with transverse wind estimator under different atmospheric turbulence conditions is analyzed.In the fourth part,the 3D registration method for AO-OCT retinal image is introduced.Fixational eye motions during AO-OCT imaging lead relative motion deformation between imaging results.The 3D registration method for AO-OCT retinal image mainly includes two parts: pre-registration and fine registration.Pre-registration reduces the search space for fine registration by sampling and phase correlation,and improves the matching efficiency.Fine registration is based on the results of preregistration,and the matching calculation is in the sub-image.This method improves the calculation speed and registration accuracy.The outlier filtering and interpolation for the fine registration results further reduces the mismatch.The AO-OCT retinal image registration results show that the sharpness ratio and the structural similarity index(SSIM)of the registered image are significantly improved compared to the unregistered image.In the fifth part,an intra-volume eye motion artifact correction algorithm for AOOCT retinal image is proposed.The AO-OCT 3D registration method is used to measure the fixational eye motions.Then the eye motions in the retinal volumetric image are estimated.Finally,the intra-volume eye motion artifact is corrected.Three metrics are introduced to measure intra-volume eye motion artifacts and evaluate correction performance.After lateral eye motion correction,the eye motion artifacts in the enface image is reduced,and the distortion of the cone image strip is reduced.After the correction of eye motion in the axial dimension,the bound of the retinal layer is smoother,the standard deviation of the cone outer segment tips(COST)segmentation depth is reduced.After correction of eye motion in all three dimensions,the decreases in the peaks at the volume rate and subsequent harmonics are observed in the power spectral density of eye motions in the axial,line,and scan dimensions.This thesis focuses on the predictive control and image registration in AO systems using motion estimation.It mainly solved the problem of time delay error in astronomical imaging AO system and eye motion artifacts in retinal imaging AO-OCT system.The research still needs future works to improve the content,such as the experiments of transverse wind estimation and predictive compensation for real atmospheric turbulence.