| Description | This project explores a semi-automatic/fully automatic approach to reconstructing UK city regions into 3D buildings. City reconstruction presents significant research challenges, particularly due to the scarcity of high-quality, annotated 3D data, such as CAD models, meshes, or wireframe representations. This project aims to address some of these challenges by investigating the following tasks, but not limited to:
1) Exploring tools to extract 3D data from various sources such as satellite images, drone imagery, and street views, leveraging platforms like Google Maps/OpenStreetMap.
2) Developing user-friendly tools or scripts for generating wireframe or editable polygonal representations of UK houses from point cloud data, integrating existing tools where possible to enhance efficiency.
3) Scaling these techniques to process large datasets at the city level.
4) Generating accurate point cloud representations through methods such as LiDAR simulation, physical 3D model assembly, 3D printing, and photogrammetry-based reconstruction.
5) Creating high-quality, annotated datasets [(point cloud and polygonal representations) pairs] to support deep learning and machine learning applications for automated 3D building generation and reconstruction.
6) Advancing automation through deep learning, machine learning, and optimization techniques to efficiently convert point clouds into editable 3D polygonal representations at scale.
A key focus is the development of high-quality, well-annotated datasets, forming the foundation for robust deep learning and machine learning applications. The project is designed to be flexible and scalable, allowing for future expansions and customizations based on emerging challenges and technologies. |
|---|