Abstract:Building Information Modeling (BIM) and 3D laser point cloud models are widely used in power facility inspection and maintenance. However, the sampled point clouds converted from BIM models and the laser-scanned point clouds often suffer from inconsistent reference coordinate systems. To address this issue, this paper proposes an automatic registration method based on a two-stage progressive inlier update strategy and multi-feature fusion. First, the method converts the BIM model of a pylon into sampled point clouds using a sampling tool and down samples both the sampled and scanned point clouds. Next, a KD-tree-based data structure is established to accelerate neighborhood searches for point cloud normal estimation, followed by extracting Fast Point Feature Histogram (FPFH) features. Then, an initial transformation matrix is obtained using the Random Sample Consensus (RANSAC) algorithm with the two-stage progressive inlier update strategy. Finally, precise registration is achieved through an Iterative Closest Point (ICP) algorithm incorporating multi-feature fusion based on curvature and normal vectors. Experimental results demonstrate that, compared to traditional algorithms, this method improves the registration accuracy of pylon point cloud data by 6.91% while enhancing computational efficiency by 8.0%, validating its effectiveness and practicality.