Automated recognition and 3D CAD modeling of standardized steel bridge members in a laser scan

Recently, 3D models of bridges are being increasingly used for maintenance information management where the inspection results on the damages and degradations can be associated with the corresponding portions on the model. In... [ view full abstract ]

Recently, 3D models of bridges are being increasingly used for maintenance information management where the inspection results on the damages and degradations can be associated with the corresponding portions on the model. In Japan, 3D models of existing bridges are not usually provided; therefore, automated 3D modeling of bridge structures in laser scans are receiving a lot of attention. In particular, the superstructures of steel bridges include many standardized steel members, such as L-shaped angle steels; therefore, there is a great need for efficient as-built modeling of these members based on laser-scanned point clouds. The purpose of this study is to develop a fully-automatic method to recognize standardized steel members from point clouds captured by a single laser scan. First, we precisely classify the visually-feasible geometric relationships between the primary planar regions of members when observed by a single laser scan. Using our classification, various types of steel members that have arbitrary cross-sectional shapes can be recognized automatically even when only two primary planar regions on a member can be scanned and when the measurements are partially missing because of the occlusions on the observed primary regions. The algorithm can work for a broad range of standardized steel members in a uniform way, including L-shaped, CT-shaped, H-shaped, and U-shaped steel members. Moreover, the existence domain of an individual member can be automatically extracted based on the classification of the relationship. We evaluated the recognition accuracy for the point clouds of the superstructure members of a short-span beam bridge and achieved 100% precision and 90% recall.