Structured Classification Feature-Guided FPFH-Based Coarse Registration Method for Precast T-Girders

In precast girder yards, dense storage, limited scanning space, and mutual occlusion between girders make coarse registration of multi-station point clouds susceptible to noise, missing data, and insufficient overlap. To address these challenges, this study proposes an FPFH-based coarse registration method guided by structured classification features for precast prestressed concrete T-girders used in highway bridges. Based on the geometric characteristics of precast T-girders, engineering-oriented criteria are established to identify planar, linear, and corner features. Representative points from these features are then selected as FPFH descriptor centers. This introduces structural prior information into descriptor center selection while retaining the standard FPFH descriptor and SAC-IA coarse registration framework. Comparative experiments are conducted using 13 simulated Gaussian noise conditions and four local point cloud incompleteness conditions. The method is further validated using four scans of a single 30 m standard precast T-girder acquired at a precast girder yard. The results show that, compared with random-sampling FPFH, the proposed method reduces the coarse registration RMSE by 72.12% at a Gaussian noise level of 10 mm. Under local point cloud incompleteness ranging from 10% to 40%, the coarse registration RMSE is lower in all tested cases. At 40% incompleteness, the rotation errors about the three coordinate axes are reduced by more than 85% on average. The average runtime in the Gaussian noise simulations is reduced by 22.42%. For the field data, the average coarse registration RMSE and runtime are reduced by 62.42% and 28%, respectively. After ICP refinement, the average registration residual RMSE reaches 2.12 mm, representing an 80.97% reduction compared with random-sampling FPFH. These results indicate that the proposed method achieves favorable coarse registration performance under the tested conditions. It may provide a technical reference for multi-station point cloud merging and data preparation for subsequent geometric quality inspection of precast T-girders.