Bundle adjustment for multi-source Mars orbiter imagery with generalized control constraints

Abstract

Integration of multi-source Mars orbiter imagery leverages diverse datasets to enhance surface details and mapping accuracy. Achieving spatial consistency is crucial, yet challenges arise from significant resolution differences and the lack of ground control points. This paper proposes a bundle adjustment method with generalized control constraints tailored for multi-source orbiter imagery. A bias compensation model using affine and cubic spline functions is employed to effectively correct linear and nonlinear distortions. Generalized control constraints integrate planar and elevation aspects through planar control points, virtual control points, tie points, and terrain information. Planar weak control constraints are established by crater features extraction utilizing the Segment Anything Model 2, ellipse fitting, and accurate point set matching, with further enhancement from slope information derived from reference terrain. Elevation control constraints are established using the reference Digital Elevation Models and slope data. An adaptive weighting strategy is developed to combine these aspects optimally, facilitating precise registration of multi-source orbiter imagery to a unified Mars reference frame. Experiments with Mars Context Camera (CTX) and High-Resolution Imaging Science Experiment data validate the performance of the proposed method across various Martian terrains, demonstrating significant reductions in reprojection residuals from approximately 10 pixels to under 0.8 pixels. Additionally, the planar positioning accuracy improved from a maximum of over 300 m to within 18 m, with elevation accuracy within 3 m. Comparisons with existing adjustment methods and fully controlled CTX mosaic products confirm the effectiveness and reliability of the proposed method.