Geometric Accuracy Assessment of Tie Point-Based RFM Refinement Using Bundle Adjustment Framework

초록

Rational Function Model (RFM) are widely used for satellite image georeferencing. However, initial RFM parameters often contain geolocation errors due to orbit, attitude, and sensor modeling limitations. This study presents a bundle adjustment framework that refines RFM using tie points without external ground control points (GCPs). Multi-view experiments with two to seven images demonstrated that increased redundancy enhanced adjustment stability. The adjustment operated in a relative model space, and its effectiveness was validated using GCPs. Results showed that reprojection errors of tie points remained below 0.6 pixels, while independent check points showed errors under 0.8 pixels. Evaluation with GCPs confirmed reductions of 1-5 pixels in image-space errors compared to the original RFM. Furthermore, residual error patterns revealed systematic trends, suggesting potential for further correction. These findings indicate that referencing imagery with minimal GCPs could anchor the refined model to absolute coordinates, thereby reducing GCP dependence in large-scale correction workflows. © ACRS 2025.All rights reserved.

키워드