|Supervisor:||Prof. Gudrun Klinker|
|Advisor:||Rudolph, Linda (@ge29tuw)|
|Submission Date:||30th June 2022|
Monitoring and documenting the construction progress of buildings is an essential task. A major part of this is the comparison between the as-planned model and the as-built state of a construction site. The main goal of this is to detect deviations from the planned model as early as possible. These deviations provide information whether adjustments are necessary and therefore have to be measured and visualized. Being currently still a mainly manual and time-consuming task, automation in this area is quite beneficial and indispensable in the future. In this thesis, we provide methods to overcome these challenges and automate most of the subtasks of this process. For this, our pipeline starts with the acquisition of the necessary scan data. Either photogrammetric methods or laser scanners are used to collect these scans as 3D point clouds. Furthermore, to obtain the data of the planned building information models, we need to work with Industry Foundation Class (IFC) files. In the context of data acquisition and processing, we support to work with a wide variation of different 3D point cloud and mesh formats. When only comparing parts of a building, these components need to be extracted and assigned. We propose using a state-of-theart Neural Network for semantic segmentation or classic plane segmentation. In order to properly prepare the data for the next step, further preprocessing is needed which is implemented fully automated in this project. This data is then used for aligning the scan with the model which is the main focus of this thesis called registration. The registered results are then evaluated with different numerical quality metrics and also visualized for verification of the construction progress. For multiple construction states belonging to a building project, several registrations are performed. We propose a building progress visualizer to simultaneously display all of the results of these registrations for the progress analysis and documentation. All of our methods can also be called via an intuitive and interactive User Interface.
[ PDF (optional) ]
[ Slides Final_MT_Presentation_Xiao_Elisa]