Semester Award Granted

Spring 2025

Submission Date

May 2025

Document Type

Thesis

Degree Name

Master of Science (MS)

Thesis/Dissertation Advisor [Chair]

Sudhagar Nagarajan

Abstract

3D Building reconstruction is crucial for urban planning, enabling stakeholders to make informed decisions on critical issues such as flood risk assessment, urban heat island effects and sustainable infrastructure development. Accurate classification of LiDAR point cloud data is fundamental for identifying building structures and separating them from other elements in complex urban environments. In this study, a deep learning-based approach, RandLA-Net, is employed to classify building points from airborne LiDAR data. This classification process serves as a foundation for generating reliable 3D building models. To enhance classification accuracy, this research integrates building footprints as a reference layer, refining the detection of building points and ensuring the validation of the classification results. By aligning classified building points with derived building footprints, the study improves the precision of extracted building points. Once the building points are improved, they are utilized to reconstruct detailed 3D building geometric models at different levels of detail including LOD 0, 1 and 2. This research further focuses on reconstructing and automating the workflow for LOD extraction by employing the Open3D library to generate LOD models efficiently. By combining advanced classification and building footprint extraction techniques with an automated building reconstruction process, this approach optimizes the utility of LiDAR point cloud data, providing detailed LOD models that support a wide range of spatial decision-making processes.

Download

Share

COinS