Modelling And Compensation For Attitude Jitter Of Multi-view High-resolution Satellite Images

In space photogrammetry,the accuracy of geo-positioning is one of crucial factors which determine the accuracy of mapping based on high-resolution satellite images.Attitude jitter is a major error source restricting the geo-positioning accuracy.A lot of high-resolution satellite images are suffering from attitude jitter.In order to improve geopositioning accuracy,the research on the methodology of detection and compensation of attitude jitter is very necessary.This article introduces high-accuracy external DSM as reference data and explores the methodology for detection and compensation of attitude jitter from the aspects of attitude model,multi-view images,detection and compensation for height errors:(1)The methodology of detection and compensation for attitude jitter based on the penalized spline attitude model is proposed.In the stage of producing sensor-corrected product from original CCD images through the sensor correction,an appropriate attitude model is chosen to absorb errors caused by attitude jitter.However,the attitude models that exist cannot meet the requirement of high-accuracy modelling of attitude.Thus,a penalized spline model is proposed,and Lagrange,cubic polynomial and Chebyshev model are introduced as comparison.The method for assessment based on the relative and absolute height accuracy of DSM derived from stereo image pairs is proposed to evaluate the ability of attitude modelling of attitude models: calculating the height difference between forward-nadir and backward-nadir DSM;calculating the height difference between forward-backward DSM and referential DSM,evaluating according to height difference and its distribution.The original CCD data of ZY-301 satellite is taken as an example to experiment and validate.The relative and absolute height accuracy of the DSM based on the penalized spline model are both the highest.Thus,the penalized spline model has the strongest ability to model the attitude jitter,absorbing the maximum amount of attitude jitter.The other three models cannot model attitude jitter correctly.(2)The detection and compensation for attitude jitter based on the adjustment of dense tie points of multi-view images.This method aims at the attitude jitter that attitude sensors haven’t detected because of the undersampling and so on.Numerous dense tie points are extracted based on multi-view images.Dense tie points are taken into the self-calibration bundle adjustment model to adjust.Attitude jitter can be spread into the image space residual errors through self-calibration bundle adjustment according to the error propagation theory.So,the compensation model for attitude jitter can be established based on image space residual errors.Attitude jitter-caused image space errors are fitted based on cubic B-spline.Attitude jitter compensation model in the image space is established and introduced into the imaging model and self-calibration bundle adjustment is implemented again to validate the effect of attitude compensation.This paper utilizes 5 scenes of World View-3 images to implement experiment.After the compensation for jitter,the image residual errors of 5 scenes are all reduced.The RMSEs of line and sample coordinates reduce 0.172 and0.349 pixels at most,respectively.The jitter effect of the residuals in the line direction is eliminated.(3)The compensation for attitude of high-resolution images based on the referential digital surface model.When the view amount of the multiview images is insufficient,for example,if there is only one dual-view stereo image pair,the method of the compensation for jitter based on dense adjustment can only compensate for the attitude jitter in the roll angle direction but is not able to handle with the pitch angle error.Errors of pitch angle cause significant height errors of high-resolution images.Aiming at this problem,the referential DSM is registered with the DSM derived from stereo image pairs.The height errors compensation model is set up based on the height difference between two DSMs.Compensated tie points corresponding ground points are reprojected into the image space and the coordinate difference compared to original image points is calculated.The compensation model in the object space is converted to the image space.The compensation model in the image space is introduced into the imaging model to correct the attitude jitter in the image and the image without jitter distortion is generated.The stereo images of ZY-302 satellite are used in the experiment,and the height errors in images are detected.The average and RMSE of the height difference compared to the referential DSM is-0.013 and 1.943 meters respectively,after the height compensation,the height difference of the DSM is reduced to-0.002 and 1.738 meters.The height accuracy is improved.