AbstractThree‐dimensional (3D) feature‐matching techniques, which are essential for progress towards an automated feature‐based procedure, have attracted considerable research attention in both the photogrammetry and computer vision communities. This study introduces a novel matching approach, called RSTG, that comprises four major phases: rotation alignment; scale estimation; translation alignment; and geometry checks. These steps efficiently determine a feature‐based correspondence and frame transformation between datasets. RSTG analyses the similarity and relative geometry of features by employing feature observations and their uncertainty; this allows different types of features to be matched exclusively or simultaneously. This study validates the proposed method with both simulated and real datasets, demonstrating its effectiveness with satisfactory matching rates in a diverse range of feature‐based point cloud registration tasks.