With the deep integration of digital and non-destructive testing technologies, the development of high-precision 3D structural reconstruction methods has become an urgent need in the engineering field. In this study, a point-cloud-driven digital reconstruction method for structures is proposed based on laser-scanned point cloud data. The method integrates Cloth Simulation Filtering (CSF) and Pass-Through Filtering to remove noise points. After initial registration based on feature point matching, the Iterative Closest Point (ICP) algorithm is employed to achieve fine registration of multi-view point clouds. High-precision reverse modeling of the structure is realized by combining feature extraction techniques. A laboratory cantilever beam was used for geometric-mechanical dual validation. The results show that the relative errors of the first three natural frequencies of the reconstructed model are 2.815%, 8.210%, and 8.693%, respectively, while the dimensional errors range from 0.00375% to 1.23%. In conclusion, the 3D model of the cantilever beam reconstructed from point clouds exhibits high consistency with the physical model in terms of key mechanical properties (such as natural frequencies) and geometric dimensional accuracy, demonstrating its potential to replace physical models for related analyses.