Van visuele data naar semantische modellen: geautomatiseerde semantische verrijking van gebouwde omgevingen

Read full text Summarise with VIRAT
Authors: Umamaheswaran Raman Kumar
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This dissertation examines how advances in computer vision and deep learning can support the semantic enrichment of the built environment by enabling automated and scalable interpretation of spatial data. At the indoor scale, it introduces BelHouse3D, a benchmark dataset designed to capture realistic occlusion scenarios in residential interiors that are absent from existing datasets. The research also proposes a domain-specific evaluation metric for assessing the quality of indoor reconstructions and presents an enhanced reconstruction pipeline that improves the robustness and interpretability of three-dimensional indoor models. At the urban scale, the dissertation develops an end-to-end workflow for mapping cityscapes from very high-resolution aerial imagery using RGB and LiDAR data, enabling the joint classification of urban structures and materials and the integration of results into geographic information systems. The approach is validated across six Belgian cities and eighteen material classes and shows improvements compared with conventional baselines. The dissertation also introduces similarity weighted Intersection over Union, or sIoU, an evaluation metric that incorporates semantic similarity between classes and can be applied to both two-dimensional and three-dimensional segmentation tasks. Collectively, these contributions advance methods for producing more accurate and interpretable digital representations of the built environment.

Fields of Study & Subjects

Automated Road and Building Extraction 3D Surveying and Cultural Heritage Advanced Neural Network Applications

Publication details

Journal / source
Lirias
Publication date
17 March 2026
Language
EN
Document type
Thesis