Ohio State University Researchers Develop SAT2LoD2: An Open-Source Python Tool For 3D Landscape Modelling Using Satelite Imagery

This summary article is based on the paper 'SAT2LOD2: A SOFTWARE FOR AUTOMATED LOD-2 BUILDING RECONSTRUCTION
FROM SATELLITE-DERIVED ORTHOPHOTO AND DIGITAL SURFACE MODEL' and all credit goes to the authors of this paper.

Please don't forget to join our ML Subreddit

3D landscape modeling has seen a rise in its popularity and applications in recent years. It has countless applications in the fields of civil engineering, earth sciences, military applications, and many others. Geometric 3D models are typically developed using the city geography markup language (CityGML), and the Level-of-Detail (LoD) building model is the preferred model for building 3D models using CityGML. 

The use of Satellite imagery for landscape modeling provides the advantage of covering a wide area and is low cost. However, developing LoD2 models using satellite imagery remains a big challenge. Building models in such a way involves complex steps demanding heuristics-based approaches and ML-based detection paradigms.

In a recent paper, researchers at the Ohio State University propose a SAT2LoD2 to facilitate the development of 3D landscape models. SAT2LoD2 is an open-source, python-based GUI-enabled software that takes the satellite images as inputs and returns LoD2 building models as outputs. The software also has the feature of taking road networks and custom maps as additional inputs for better results.

SAT2LoD2 generates the segments of the buildings from the input image and draws a boundary for each building. Further, the software structures the roof designs based on the perimeter of the building and draws a mesh structure based on the design. Each step is independent of the other and can be independently used for other purposes as well.

The team conducts experiments using satellite imagery from the cities in the USA(Columbus), the UK(London), and Italy(Trento). The results show that SAT2LoD2 can model these urban landscapes efficiently. It can also be inferred that the quality of results from SAT2LoD2 depends largely on the quality of input image files. 

Conclusion

In a recent paper, researchers at the Ohio State University develop SAT2LoD2, open-source, GUI-based software for generating LoD2 3D landscape models. It is a python based software with a pytorch architecture aimed to provide an easy and robust method to model 3D landscapes using satellite imagery. The results indicate that the software performs the task efficiently and allows the researchers to provide additional inputs to improve results further. The team plans to improve upon the processing speeds and scalability of the software in their future endeavors with the software. 

Paper: https://arxiv.org/pdf/2204.04139v1.pdf

Github: https://github.com/gdaosu/lod2buildingmodel