Reliability-weighted fusion of multiview photogrammetric point clouds for 3D terrain reconstruction of the lunar surface

Abstract

Benefitting from advances in photogrammetry and computer vision, 3D point clouds generated from dense image matching have been proved to be an accurate, reliable, and cost-effective data source for lunar topographic mapping. To achieve a full coverage mapping of the lunar surface, a merging of point clouds generated from multiple observations is mandatory. Due to the limit of dense matching accuracy and accumulative registration errors, integrated point clouds normally suffer from disturbed stratification, outliers, and redundant points, resulting in sharp edges between the seams of the fused point clouds. To address the seaming problems of merged point clouds and achieve a seamless fusion result, a weighted fusion strategy evaluating the reliability of points from triangulation errors in the photogrammetry process is proposed. The global registration and post-processing of point clouds are also addressed to optimize the result. With comparisons to other software outputs, a DEM with a resolution of 6 m/pixel is produced, with a lower bias to ground truth and better visualization. As a result, the proposed method can improve the completeness and precision of digital surface model to a certain extent and satisfy the application requirements.