Inhaltszusammenfassung

Currently, most satellite images are geo-referenced using rational polynomial coefficients (RPCs); however, these mathematical models often fail to accurately establish the relationship between ground and image coordinates. This inconsistency can lead to shifts and drift biases in the RPCs of satellite stereo images. Such biases can significantly impact various photogrammetric applications, including displacement estimation, 3D feature extraction, epipolar resampling, and ground truth dispari...Currently, most satellite images are geo-referenced using rational polynomial coefficients (RPCs); however, these mathematical models often fail to accurately establish the relationship between ground and image coordinates. This inconsistency can lead to shifts and drift biases in the RPCs of satellite stereo images. Such biases can significantly impact various photogrammetric applications, including displacement estimation, 3D feature extraction, epipolar resampling, and ground truth disparity map estimation from two-view satellite images. Thus, correcting RPC biases is essential for achieving the precision necessary in photogrammetric mapping. This paper proposes a novel method for compensating relative RPC biases. The proposed method employs the speeded-Up Robust features (SURF) operator with a tilling strategy to extract corresponding image points. Given that these extracted points may contain outliers, the random sample consensus (RANSAC) algorithm based on the mathematical model of the RPCs is proposed to simultaneously detect the outliers and estimate the relative biases of the RPCs. Unlike the original RANSAC + RPC algorithm, the proposed progressive RANSAC + RPC algorithm accounts for both relative shift and drift biases concurrently. Additionally, this method updates the coefficients, and RPC normalization parameters without introducing additional parameters. Results from experiments on IRS P5, World View III, and ZY3 stereo pairs demonstrated that the relative RPC biases were compensated with sub-pixel accuracy. Furthermore, the findings indicate that the proposed progressive RANSAC + RPC algorithm outperforms the original RANSAC + RPC algorithm and exhibits greater effectiveness in outlier detection compared to traditional RANSAC + affine and RANSAC + projective algorithms.» weiterlesen» einklappen

Autoren

Klassifikation

DFG Fachgebiet:
Geophysik und Geodäsie

DDC Sachgruppe:
Geowissenschaften