This project is supported by NTNU Digital.

Brief Introduction

Semantic 3D building models are essential digital infrastructure in many applications such as urban planning, facility management, disaster management, energy simulation and others. Currently, many 3D building models are available as a collection of polygons representing unstructured geometry and lacking semantic meaning. While such models may still be valuable for visualisation and other purposes, their full potential in 3D GIS analyses is hindered by the lack of semantics. 

Aiming to close the abovementioned research gap and support simulation, calculation, query and interactive visualization in various applications, the objective of this project is to provide a promising digitalization solution to generate 3D building models with detailed semantic information, from roof faces to facade elements. 

Test Area

Trondheim is selected as the test site of this project. All buildings in the Trondheim area will be reconstructed in 3D models with semantic information. Through this project, Trondheim will be the first municipality in Norway to have large-scaled 3D building models in LoD3. With the support of this project, it is possible to generate nation-wide 3D buildings in high levels of detail (LoDs) for Norway. This will be a great contribution for Digitalization in Norway. 

Recent Outcomes

• Algorithms for automatic building footprint reconstruction (LoD1)
  • Generative Adversarial Network (GAN)-based method 
  • PH-Shape
• Algorithms for 3D building reconstruction with roof semantic information (LoD2)
  • Roof vertices and edges information-based method for roof reconstruction
  • Roof plane-based method for roof reconstruction
  • Large-scale 3D building reconstruction