IMPROVEMENT OF RELATIVE GEOMETRIC CORRECTION FOR MULTIPLE SATELLITE IMAGES BASED ON HOMOGRAPHY BUNDLE ADJUSTMENT

초록

Relative geometric correction of satellite images is a technique that adjusts geometric displacements among them based on their relative positional relationships in image space. It has a vital role in improving the overall quality, precision, and reliability of satellite image analysis. With advancements in satellite imaging technology, high-resolution and high-quality images are being collected daily from satellites deployed worldwide. However, in spite of advanced satellite geo-positioning equipment, there are still diverse levels of remaining positional errors. These errors need to be corrected through geometric correction. In case of utilizing more than two satellite images over the same geolocation, performing relative geometric correction among them and an image-to-image registration is still challenging. The challenges lie on selecting rational reference image, or reference plane, on determining image transformation and achieving optimal solution for all image combinations. In this study, we propose a homography-based bundle adjustment for relative geometric correction of multiple satellite images. Our objective is to estimate the optimal reference plane onto which the images are projected and to achieve a fast generation of result images. For experiments, we used multiple high-resolution satellite images from KOMPSAT-3A. The processing level of the used satellite images was L2G, which was processed with precise orthogonal correction. We achieved relative geometric correction of multiple satellite images by the transform estimated by our method. As a result, an initial reprojection error of around 2.52 pixels was corrected to 1.42 pixels. We also confirmed that our proposed method showed more accurate result than a homography-based image-to-image registration method. © 2023 ACRS. All Rights Reserved.

키워드