A skeleton-based architectural point cloud fusion method was explored to address the challenges of model incompleteness in real-world 3D architectural models. The process begins with reverse modeling of the architectural scene using 3D point clouds. The acquired reverse point cloud data was combined with the original architectural design data to generate a forward point cloud model. A method based on the rotational symmetry axis (ROSA) was then applied to extract skeleton lines from both the reverse and forward point cloud models. The fusion of the forward and reverse point cloud models was achieved by skeleton matching, resulting in a reconstructed 3D model with improved completeness. Experimental validation shows that this method significantly reduces areas of model incompleteness, providing new insights and methods for 3D modeling and reverse engineering in architectural reality capture.