Document Type : Original Research Paper
Authors
1 Department of Civil Engineering, Faculty of Engineering, Bu-Ali Sina University, Hamadan, Iran
2 Department of Geomatics Engineering, Faculty of Civil Engineering and Transportation, University of Isfahan, Isfahan, Iran
Abstract
Background and Objectives: In recent decades, geomatics science has made significant progress, and these advances are due to advanced measurement tools and innovative technologies in the field of geometric and spatial data acquisition. In this context, portable laser scanners and UAVs have been introduced as basic and efficient tools that are capable of accurately and quickly measuring various objects and environments, including urban spaces. These devices automatically record all the details of the urban space in the form of point clouds or images. To extract the geometric information of buildings from these details, it is necessary to use machine vision methods. To achieve accurate and reliable models of buildings, a sequence of post-processing operations is implemented when processing point cloud data. One of the most important stages of these processes is the segmentation of the point cloud. These steps transform point cloud data into more conceptual and analyzable information. One of the important issues in processing point cloud data is the ability to extract flat surfaces of building facades (walls). These flat surfaces are of special importance as basic components in modeling and analyzing the condition of buildings. Accuracy in the information related to these flat surfaces allows for a more accurate and complete distinction between different components of buildings. This is important in several applications including urban planning, construction management, and energy consumption analysis of buildings.
Methods: In this article, the combination of MSAC and G-DBSCAN algorithms is used to extract flat surfaces from three-point cloud datasets (point cloud obtained from GeoSLAM ZEB-HORIZON laser scanner devices, point cloud obtained from Phantom 4 Pro drone imaging and hybrid point cloud) has been These two algorithms are executed sequentially. The area chosen for this purpose is the buildings of the Faculty of Engineering of Bu-Ali Sina University in Hamedan. Because this environment has features such as architectural diversity, the existence of flat facades, and different ways of placing walls in relation to each other with different dimensions.
Findings: This research, with a comprehensive evaluation of three separate data sets, shows an average precision of more than 97%, which guarantees high accuracy in data extraction. In addition, the average recall has reached more than 94%, which covers most of the elements of the facade. The result of this evaluation is the F1 score with an average of 95%, which indicates progress in the field of accurate building data extraction and architectural modeling. However, the algorithm encountered challenges when facing the walls that were perpendicular to the laser scanner's movement path, which reduced the representation rate. Also, the SfM algorithm has difficulty in generating points on window panes, which caused some points related to the space inside the windows to be recognized as wall points. This issue shows that point cloud generation algorithms from images affect the results of this algorithm. On the contrary, the results of the combined data have been very promising, in such a way that these data converged faster than the other two data sets in the first step of the algorithm and had high performance in Precision and Recall.
Conclusion: However, the findings show that the algorithm has generally shown an outstanding performance in extracting building facade information, especially with the use of diverse and varied data. These developments are promising and open new horizons in spatial data analysis and building modeling. This innovative approach can be used in various applications and help to develop modern and data-driven architectural models.
Keywords
- UAV photogrammetry
- Mobile laser scanner
- point cloud segmentation
- point cloud clustering
- DBSCAN and RANSAC algorithms
- Plane extraction
Main Subjects
COPYRIGHTS
© 2024 The Author(s). This is an open-access article distributed under the terms and conditions of the Creative Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) (https://creativecommons.org/licenses/by-nc/4.0/)
)