Semantic Segmentation and 3D Reconstruction of Steel Girder Bridge from TLS Point Cloud Integrating Domain Knowledge

Component-based inspection and maintenance of steel girder bridges increasingly benefit from three-dimensional as-is structural models that explicitly represent individual members and inspection elements. However, most existing bridges lack digital structural models, and generating such models from scanned point clouds remains a nontrivial task due to occlusions, missing data, and the absence of design drawings. This study proposes a heuristic method for constructing lightweight parametric models of steel girder bridges from terrestrial laser scanning (TLS) point clouds. The proposed method integrates image-based neighborhood search, point-based dimensionality analysis using principal component analysis, and line-based and plane-based RANSAC to extract and reconstruct major structural components, including deck slabs, main girders, and cross beams. By exploiting domain knowledge and the geometric characteristics of steel girder bridge superstructures, the method enables component-wise segmentation and reconstruction without relying on design drawings or prior structural models. The proposed approach was validated using real-world TLS data of an existing steel girder bridge, demonstrating stable extraction and reconstruction of major structural components. The resulting parametric models explicitly represent inspection elements and have the potential to facilitate the spatial association of inspection results, photographs, and maintenance records, thereby supporting practical bridge maintenance workflows. The applicability and limitations of the proposed method are also discussed.