Advancing scan-vs-BIM in progress monitoring of infrastructure construction

The upgrade and realignment of transport infrastructures can stimulate the local economy and benefit the area alongside the traffic system. However, the low project performance of large-scale structures remains a significant issue within the construction industry, contributing to potential cost overrun and project delay. The application of cutting-edge technologies enhances the efficiency of reality capture while also introducing new challenges in data collection and processing, particularly in the construction of large-scale transport infrastructures. As an alternative to Scan-to-BIM for construction progress monitoring, Scan-vs-BIM demonstrates outstanding performance across various construction projects, whether in indoor or outdoor complex site conditions. Compared with Scan-to-BIM, it avoids error accumulation by integrating essential steps, such as object recognition and point segmentation, directly into the comparison framework. Additionally, it reduces reliance on 3D reconstruction by introducing multiple geometric criteria. However, Scan-vs-BIM has several inherent shortcomings at both micro and macro levels. These issues include challenges related to the accuracy of point cloud registration, progress determination, and conceptual limitations inherent in the framework. This thesis aims to enhance the performance of the conventional Scan-vs-BIM framework and explore potential solutions to address its conceptual limitations in construction progress monitoring. It integrates Hausdorff distance and a classification-based surface repair method to handle misalignments and occlusions in datasets collected via TLS. Results from the repaired point cloud demonstrate maximum errors of 4.5% for 3D reconstruction and 2.7% for 2D surface calculations. To further mitigate occlusions, eliminate misalignments, and improve efficiency, a handheld SLAM-based laser scanner was introduced, accompanied by a compatible framework. The proposed Scan-vs-BIM framework can complete the scanning of a 100-meter segment of an overpass within 4 minutes, achieving a minimum accuracy of 73.6%, even in the presence of temporary structures and external factors contributing to severe occlusions in the raw data. Additionally, based on G4PCS, a RoLFC keypoint extraction method and the accompanying SiG4PCS algorithm were developed to automatically register 3D BIM models to as-built point clouds, achieving maximum rotation and translation errors of 1.06° and 0.63 m, respectively. Finally, an automated framework for extracting road network construction sites was proposed. With a minimum accuracy of 94.50%, the method can effectively support Scan-to-BIM through parametric modeling, offering a potential solution to address the conceptual limitations of Scan-vs-BIM. This thesis enhances the performance and extends the functionality of Scan-vs-BIM construction progress monitoring framework through several modifications. The main academic contributions can be summarized in three key aspects: (1) Enhancing the performance of original Scan-vs-BIM frameworks by addressing data quality issues such as occlusions and misalignments, enabling fully automated progress determination through reasonably selected geometric criteria, and improving data collection efficiency; (2) Developing a novel RoLFC key point extraction method and a compatible SiG4PCS to enable accurate and automatic registration in Scan-vs-BIM; and (3) Proposing a prototype of automated construction site extraction framework for Scan-to-BIM progress monitoring in road construction. The results demonstrate that selecting an appropriate laser scanning platform is crucial for effective progress tracking, and proposed occlusion repair method is essential for fully automating progress determination facing the data quality issues. Furthermore, the study confirms that automatic point cloud registration can be achieved at the point level rather than the plane level, even in cases of significant geometric deviation between datasets. The automatic extraction of road construction sites offers potential fo integrating Scan-vs-BIM and Scan-to-BIM approaches, thereby extending their functionality and highlighting their significant value in industrial applications.