Highly Accurate and Automated Point Cloud Registration for Full-Field Deformation Measurement and Construction Quality Inspection of a Steel-Concrete Composite Bridge Pier

Steel-concrete bridges have been widely constructed within road networks due to their distinctive mechanical properties. Nevertheless, issues related to construction quality, such as concrete pouring blockages, steel plate delamination, and steel plate deformation, frequently occur, posing significant risks to the safe operation of these bridges. Laser scanning technology provides an effective solution for rapid construction quality inspection of bridges, owing to its rich information, high data density, and noncontact measurement capabilities. However, conventional data processing methods are hindered by cumbersome procedures and limited accuracy in point cloud registration for deformation extraction. To address these challenges, this study developed a highly accurate and automated full-field deformation measurement method for the construction quality inspection of steel-concrete composite bridge piers using laser scanning technology and deep learning. The innovative contributions of this study are twofold: (1) a geometric feature-based point cloud simplification algorithm was developed to automatically select sampling points while preserving crucial structural information and reducing the data volume by employing the Fast Point Feature Histogram (FPFH) feature and an improved voxel down-sampling technique; (2) a highly accurate point cloud registration method was devised, in which geometric structure embedding and an attention mechanism were utilized to extract point cloud features, and maximal clique constraints were applied to estimate the positional attitude between two point cloud datasets for full-field deformation extraction and construction quality inspection. The effectiveness and robustness of the proposed method were validated using public point cloud datasets and field test data from a steel-concrete composite bridge pier. Results demonstrate that the proposed method surpasses existing feature-based registration methods regarding translation error, inlier ratio, and root mean square error. Furthermore, the maximum deformations on the inner and outer sides of the bridge pier were measured at 8.03 cm and 8.65 cm, respectively. The average relative error of the measured deformations was found to be less than 3% when compared to the results obtained with the hand-held laser scanner. Compared with existing deformation measurement methods, the proposed method can realise bridge deformation measurements at millimeter level, and has the advantages of non-contact, high accuracy and automation, as it does not require the installation of sensors and can process the point cloud data end-to-end, with less human intervention and low technical requirements for operators.