Epipolar resampling for Satellite Linear Pushbroom Images Using Block-wise Affine Transformation Parallel Method

Abstract Epipolar resampling is an essential step for 3D reconstruction and Digital Surface Moeld(DSM) generation from satellite images. However, with improved satellite image resolutions and larger image size, the time required for epipolar resampling increases exponentially. While the polynomial fitting methods based on Piecewise Projection Trajectory Method (PPTM) is commonly employed for epipolar resampling, its effectiveness and efficiency diminish when confronted with large and high-resolution images. To tackle this challenge, we propose a novel parallel block-wise epipolar resampling method designed to expedite the resampling process without compromising accuracy. This method leverages PPTM and fixed elevation plane to establish the relationship between left and right epipolar points. Local affine transformations and image partitioning replace polynomial transformations applied across the entire image to approximate the correspondence between original and epipolar images. Furthermore, parallel computation was employed for block-wise pixel resampling acceleration. Experimental analysis using IKONOS-2, ZY-3, and GF-7 images confirms the efficacy and accuracy of our method. We achieve subpixel y-disparities comparable to polynomial fitting methods, while reducing resampling time by 10 to 20 percent through single-core serialization. Moreover, multi-core parallelism approach achieves a parallel efficiency exceeding 80%.