Prior information guided cost optimization for terrain generation from high-resolution orbiter imagery of lunar south pole

Precise reconstruction of lunar south pole’s terrain is crucial for future exploration missions, particularly for landing site selection, resource identification, and scientific research. However, the challenging illumination conditions, characterized by extensive shadowed areas, pose significant challenges to traditional photogrammetric methods. The shadowed areas lead to poor disparity estimation and less accurate terrain models, highlighting the need for improved techniques that can effectively address these difficult conditions. This study proposes an enhanced dense matching method for terrain reconstruction at lunar south pole, which incorporates prior information of existing lower-resolution terrain model to guide the semi-global matching (SGM) process. We begin by adaptively selecting seed points in shadowed areas of the image through intensity histogram analysis. Homogeneous pixels with close spatial distance and small intensity difference surrounding each seed point are then identified based on distance and intensity differences. After obtaining prior disparity values, the corresponding costs are simulated and combined with matching costs from the original 2D disparity search SGM algorithm, optimizing pixel costs to refine the cost volume and improve the robustness of the matching process. Experiments were conducted in representative areas of lunar south pole to validate the effectiveness of our method. Additionally, the performance of our method is compared with SGM variants and two advanced cost volume optimization methods, further demonstrating its effectiveness. Finally, we apply our method to several pairs of LROC NAC images to generate detailed terrain models at lunar south pole, quantitatively validating its reliability through comparisons with reference terrain data and LOLA laser points. • Enhance matching robustness in shadowed areas using prior information. • Select seed points adaptively in difficult areas to overcome random sampling limits. • Improve photogrammetric terrain mapping performance at lunar south pole.