Abstract: The lunar South Pole region’s limited illumination, extensive shadowed regions, and homogenous surface features with weak textures pose significant challenges to stereoscopic image matching and 3D terrain reconstruction. To address these issues, an Efficient Confidence-guided Stereo Matching (ECSM) algorithm was proposed. This algorithm improved matching precision and efficiency by assessing the confidence of non-support points, updating the support point dataset, constructing a triangulated mesh, and recalculating disparities within triangle vertices based on their confidence levels. On this basis, a photogrammetric method for lunar 3D terrain reconstruction was established. High-resolution image data from the Lunar Reconnaissance Orbiter’s Narrow Angle Camera was utilized for validation experiments conducted in the vicinity of the Shackleton crater within the lunar South Pole region. Qualitative and quantitative analyses of disparity maps and Digital Elevation Model (DEM) generated from different stereo matching algorithms demonstrated the reliability of the proposed algorithm in regions with weak and repetitive textures. Comparative analyses with the German Aerospace Center’s DEM and NASA’s Lunar Orbiter Laser Altimeter DEM (LDEM) for the same region revealed significant consistency in elevation and slope information, affirming the practicality and effectiveness of the proposed method. This study provides a methodological foundation for landing site selection for lunar South Pole explorations.