Building structural characterization using mobile terrestrial point cloud for flood risk anticipation
|Advisor:||Daniel, Sylvie; Badard, Thierry|
|Abstract:||Given the high frequency and major impact of floods, decision-makers, stakeholders in municipalities and public security ministry are in the urgent need to have tools allowing to predict or assess the significance of flood events and their impact on the population. Based on statistics, the first floor of the buildings, as well as the lower openings, are more likely subject to potential damage during a flood event. Thus, in the context of flood impact assessment, it would be required identifying the location of the buildings' lowest opening and especially its height above the ground. The capacity to characterize building with a relevant level of detail depends on the data sources used for the modeling. Different sources of data have been employed to characterize buildings' façade and openings. Mobile Laser Scanning (MLS) system mounted on a vehicle has proved to be one of the most reliable sources in this domain. It can produce millions of 3D georeferenced points with sufficient level of detail of the building facades and its openings, due to its street-view and close-range distance. Moreover, the increase of MLS providers and acquisitions in towns and rural environments, makes it possible to develop approaches to characterize residential houses at a provincial scale. Although being effective, several challenges are associated with extracting descriptive information of building facades using MLS data. The presence of occlusion in front of a facade makes it impossible to obtain the 3D points of the covered parts of the facade. Given the fact that windows mostly consist of glass and laser signals could not be reflected from the glass, limited points are usually available for windows. While the repetitive pattern and symmetrical positions of the openings on the facade makes it easier for the detection system to extract them, this characteristic is missing on the facade on rural and residential houses. The inconsistency of the point density in MLS data make the detection process even harder when working at city scale. Accordingly, the main objective of this research is to design and develop a comprehensive approach that effectively extracts facade openings. In order to meet the research project objective, the proposed solution consists of three phases including facade extraction, opening detection, and occlusion recognition. The first phase employs an adaptive region growing segmentation approach to extract the 3D bounding box of the facade. The second phase combines a hole-based assumption with an XZ gridding technique to extract 2D bounding boxes of the openings. The last phase which recognizes holes related to the occlusion from the openings is currently being completed. Qualitative and quantitative evaluations were performed using a real-word dataset provided by Jakarto Cartographie 3D inc. of the Quebec Province, Canada. Statistics revealed that the proposed approach could obtain good performance rates despite the complexity of the dataset, representative of the data acquired in real situations. Challenges regarding facade's self-occlusion and the presence of large surrounding occlusions should be further investigated for obtaining more accurate opening information on the facade.|
|Document Type:||Mémoire de maîtrise|
|Open Access Date:||9 May 2022|
|Collection:||Thèses et mémoires|
All documents in CorpusUL are protected by Copyright Act of Canada.