Extraction of surveying goals from point clouds obtained from laser scanners to support bridge inspection

Laser scanning enables collections of dense point clouds in minutes with mm-accuracy. Based on detailed and accurate 3D data, engineers can construct 3D as-built models and manually conduct "virtual surveying" on these models to obtain geometric information items required for their domain applications, such as "areas of rooms". This research named these geometric information items as "surveying goals".;While delivering rich geometric information, dense point clouds pose challenges for geometric information retrieval. Generally, the manual virtual surveying approach has three limitations. First, sometimes, extracting a surveying goal involves large number of similar manual measurements (e.g., multiple cross-sections of piers), which are time-consuming. Second, it is difficult to understand the accuracy and reliability of a manually extracted surveying goal. Third, the technical specifications of scanners only specify point-accuracy of the data, but additional factors, such as data density and data artifacts, can influence the accuracy of some geometric primitives extracted from point clouds and the accuracy of surveying goals related to these geometric primitives, while quantitative investigations about these factors' impacts are still limited.;Focusing on the surveying goals required by the National Bridge Inventory (NBI) program, this research addresses these limitations by conducting research studies in two directions: (1) Providing automated support for the process of extracting surveying goals from point clouds; (2) Characterizing the accuracy of two critical geometric primitives; spatial discontinuity edge and surface defect region, which have substantial impacts on the accuracy of many surveying goals. In the first direction, I formalized data processing workflows for extracting a variety of NBI related surveying goals, developed reasoning mechanisms to support automated composition and execution of computer-interpretable workflows. Results showed that this automated approach supports compositions of workflows for extracting all NBI surveying goals, and enables the evaluations of the accuracy and reliability of obtained surveying goal results. In the second direction, I developed a quantitative model for predicting the accuracy of the spatial discontinuity edge detection, and validated that model through controlled experiments and field tests. In addition, I conducted experimental studies to characterize three scanners and three algorithms for extracting surface deviating regions from point clouds.