Inhaltszusammenfassung

Historically, three-dimensional (3D) geospatial data were mainly used for visualization, while two-dimensional (2D) data underpinned spatial analyses. With recent advances in sensing and computation, as well as the demands of smart-city applications, 3D urban data, especially for building objects, has become a valuable source for processing and interpretation. This study proposes a deep-learning approach that exploits neighborhood relations on urban triangle meshes to reconstruct simplified 3...Historically, three-dimensional (3D) geospatial data were mainly used for visualization, while two-dimensional (2D) data underpinned spatial analyses. With recent advances in sensing and computation, as well as the demands of smart-city applications, 3D urban data, especially for building objects, has become a valuable source for processing and interpretation. This study proposes a deep-learning approach that exploits neighborhood relations on urban triangle meshes to reconstruct simplified 3D building models. Neural networks help overcome mesh-specific challenges such as irregular connectivity and heterogeneous sampling, enabling robust geometric analysis. Experimental results indicate that the proposed network can effectively leverage mesh data for building reconstruction. Furthermore, the results suggest that additional spatial analyses can be performed directly on mesh data, enabling the production of high-quality products. Overall, the approach delivers accurate building abstractions from real-world meshes, reduces manual post-editing, and supports downstream urban analytics. The findings highlight the growing potential of mesh-native learning for scalable 3D city modeling.» weiterlesen» einklappen

Autoren

Klassifikation

DFG Fachgebiet:
3.43-02 - Geodäsie, Photogrammetrie, Fernerkundung, Geoinformatik, Kartographie

DDC Sachgruppe:
Geowissenschaften