|Timeline:||October 2021 - 1 October 2025|
3D city model has gained attention in applications including urban planning, civil engineering, and virtual reality. With the rising demand for high level-of-detail building models, a new method to automatically reconstruct LoD3 building models has to be developed. This will also require new input data which can present the accurate structure of the buildings. We utilize UAVs to autonomously acquire image data of the buildings due to their flexibility and mobility.
Traditionally, UAV flight planning for building structure acquisition and building modeling are two separate studies. Previous flight planning research for 3D reconstruction mainly stopped at the stage of point cloud generation, generating building models from point clouds remained a separate domain. In this PhD research, we aim to directly connect flight planning and building modeling together. The generated point clouds through dense image matching will be used in reconstructing building models. The flight plan will then be adjusted automatically based on the correctness of the reconstructed model, in order to collect additional data for correcting the erroneous parts of the building models.
In the research, we will first design a nadir-view flight planning algorithm to autonomously acquire additional image data for correcting the existing LoD2 building models in a city. We will then move to develop a flight planning algorithm to autonomously acquire image data for LoD3 building modeling. This may be done by either developing a sampling and optimization approach or a search-based next-best-view approach (preferred) for deciding the optimal camera viewpoints and flight trajectory. On the other hand, we will also develop a new algorithm for LoD3 building modeling based on a generic topological graph method. We aim at achieving an optimal autonomous flight planning algorithm specifically designed for city modeling, and an innovative building modeling algorithm that is able to generate highly complex building models in detail from point cloud data.