RPC Model Refinement via 3D Corrective Rotation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current 3D reconstruction methods from satellite imagery face inaccuracies due to measurement errors in satellite geopositioning, leading to systematic errors in triangulation and reduced accuracy of 3D models, particularly with the Rational Polynomial Camera (RPC) model, which is complex and prone to errors without sufficient Ground Control Points.
Innovation Solution
A computer-implemented method for refining RPC models through bundle adjustment, which involves determining corrected projection functions by applying a 3D corrective rotation to the original RPC models, minimizing reprojection errors between key points and their projections, and using an optimization problem to iteratively refine the RPC models until specific cost function thresholds are met.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If RPC models are used for 3D reconstruction from satellite imagery, then the reconstruction can be performed without Ground Control Points, but measurement errors in satellite geopositioning cause systematic errors in triangulation and reduce accuracy
Solution Approach 1:
The patent applies preliminary action by performing bundle adjustment and RPC refinement before the actual 3D reconstruction process. The method pre-corrects the RPC models by optimizing camera parameters and minimizing reprojection errors using available tie points, thereby eliminating systematic errors before triangulation occurs. This preliminary correction ensures that subsequent 3D reconstruction achieves high accuracy without requiring Ground Control Points.
Solution Approach 2:
The patent implements feedback through an iterative optimization process that minimizes reprojection errors. The bundle adjustment algorithm repeatedly refines camera parameters and RPC coefficients by comparing projected tie points with observed image coordinates, using the error feedback to continuously improve the accuracy of the RPC models until convergence is achieved.
2Ease of manufacture
If direct RPC correction methods are used to explicitly modify RPC coefficients, then the correction can be applied directly, but the method demands a larger number of tie points and shows poorer stability and accuracy
Solution Approach 1:
The patent applies segmentation by separating the RPC correction process into distinct components: tie point detection, bundle adjustment for camera parameter optimization, and RPC coefficient refinement. This segmented approach allows each component to be optimized independently, improving overall stability and accuracy while reducing the number of tie points required compared to direct correction methods that attempt to modify all RPC coefficients simultaneously.
3Measurement precision
If bundle adjustment strategies are used to minimize reprojection error, then the standard solution for RPC correction is achieved, but the computational complexity increases
Solution Approach 1:
The patent applies partial action by focusing the bundle adjustment optimization on the most critical parameters: camera position, orientation, and a limited set of RPC coefficients. Rather than optimizing all 40 RPC coefficients simultaneously, the method selectively refines only those parameters that have the greatest impact on triangulation accuracy, thereby reducing computational complexity while maintaining high measurement precision.
Data Source
AI summary
The invention relates to computer-implemented method for the 3D reconstruction of a ground surface area by stereophotogrammetry, comprising the steps of:determining corrected Rational Polynomial Camera, RPC, models by performing bundle adjustment (BA) of original RPC models each provided with an image of a set of images of the ground surface area acquired by a remote imaging sensor and each associated to a corresponding original projection function ({Pm}) from a 3D object space to a 2D image space, wherein determining the corrected RPC models comprises determining corrected projection functions ({Pmcor}) from the 3D object space to the 2D image space; anddetermining (PC) a 3D point cloud representative (3DPC) of the ground surface area by triangulation, based on the corrected RPC models, of stereo correspondences within images of the set of images.In accordance with the invention, determining the corrected projection functions comprises determining, for each of the original projection function, a 3D corrective rotation around a remote imaging sensor center to be applied in the 3D space before performing the original projection function.