Sat2lod2: A Software For Automated Lod-2 Modeling From Satellite-Derived Orthophoto And Digital Surface Model (2204.04139v1)

Abstract: Deriving LoD2 models from orthophoto and digital surface models (DSM) reconstructed from satellite images is a challenging task. Existing solutions are mostly system approaches that require complicated step-wise processes, including not only heuristic geometric operations, but also high-level steps such as machine learning-based semantic segmentation and building detection. Here in this paper, we describe an open-source tool, called SAT2LOD2, built based on a minorly modified version of our recently published work. SAT2LoD2 is a fully open-source and GUI (Graphics User Interface) based software, coded in Python, which takes an orthophoto and DSM as inputs, and outputs individual building models, and it can additionally take road network shapefiles, and customized classification maps to further improve the reconstruction results. We further improve the robustness of the method by 1) intergrading building segmentation based on HRNetV2 into our software; and 2) having implemented a decision strategy to identify complex buildings and directly generate mesh to avoid erroneous LoD2 reconstruction from a system point of view. The software can process a moderate level of data (around 5000*5000 size of orthophoto and DSM) using a PC with a graphics card supporting CUDA. Furthermore, the GUI is self-contained and stores the intermediate processing results facilitating researchers to learn the process easily and reuse intermediate files as needed. The updated codes and software are available under this GitHub page: https://github.com/GDAOSU/LOD2BuildingModel.

• The paper introduces SAT2LOD2, an open-source algorithm that automates LoD-2 building model reconstruction through deep learning and grid-based polygon decomposition.
• It integrates HRNetV2 segmentation and adaptive polygon extraction to accurately capture complex urban building geometries.
• Experimental evaluations in cities like Columbus and London validate its reconstruction accuracy, highlighting the critical role of high-quality DSM inputs.

Overview of SAT2LOD2: Automating LoD-2 Building Reconstruction from Satellite Data

The paper presents SAT2LOD2, an open-source software designed to automate the reconstruction of Level-of-Detail 2 (LoD-2) building models utilizing orthophoto and digital surface models (DSM) derived from satellite images. This research addresses prevailing challenges in automated 3D building model generation, offering a streamlined solution conducive to wide adoption within the geospatial data analytics community.

SAT2LOD2 is developed with a focus on integrating recent advancements in building segmentation and reconstruction, emphasizing the tool's open-source nature and ease of use. Coded in Python with a GUI interface, SAT2LOD2 incorporates deep learning-based building segmentation via the HRNetV2 algorithm, adaptive polygon extraction, and grid-based decomposition to achieve precise and detailed building models. The software supports the processing of moderately sized data sets on systems equipped with CUDA-enabled GPUs.

Methodological Contributions

1. Building Segmentation: The software leverages HRNetV2 for high-accuracy building segmentation from input orthophotos. This algorithm is adept at handling urban patterns and has been trained on a diverse dataset featuring satellite and aerial imagery, achieving substantial efficacy across varied geographic contexts.
2. Polygonal Representation and Decomposition: An innovative approach involving initial line extraction and adjustment followed by grid-based rectangle decomposition is employed. This ensures that complex building structures can be efficiently represented using simpler geometric forms, essential for the robust reconstruction of varied building typologies.
3. 3D Model Fitting and Export: After 2D building parameters are delineated, the software applies a comprehensive fitting process to generate 3D roof models, accommodating a variety of roof geometries such as flat, gable, and mansard. The transformation of building models into mesh format facilitates compatibility with further analysis or integration within broader urban modeling workflows.
4. Orientation Refinement with External Data: The methodology extends to the refinement of building orientations using OpenStreetMap road networks, reducing potential errors from DSM noise and enhancing alignment consistency within broader urban landscapes.

Evaluation and Results

Experiments conducted in diverse urban settings, including cities like Columbus and London, highlight SAT2LOD2's effectiveness in reconstructing 3D models contingent upon the quality of input DSM and building segmentation maps. The research illustrates a significant dependency on these inputs, underscoring the software’s reliance on high-quality segmentation achieved through utilizing deep learning models like UNet, HRNet, and Swin-transformer.

Quantitative analyses using metrics such as 2D and 3D Intersection over Union (IOU) validate the accuracy and efficacy of the reconstructions. Notably, the quality of DSM inputs markedly influences the geometric fidelity of the outputs, with improved DSM resolution correlating with enhanced reconstruction accuracy.

Implications and Future Directions

The development of SAT2LOD2 marks a meaningful contribution to the field of automated urban modeling. By minimizing the constraint of complex proprietary systems and enabling easy access through its open-source platform, it democratizes the ability to generate high-detail urban models from satellite data.

The paper suggests potential advancements in model library expansion and processing optimizations. Future iterations might incorporate a more extensive array of building typologies or leverage parallel processing techniques to expedite computation, thus broadening the software's applicability and efficiency.

In summary, SAT2LOD2 serves as a valuable asset to the geospatial research domain, providing a practical, effective, and accessible tool for advancing urban spatial analytics. Its open-source nature encourages collaboration and innovation, setting a precedent for further developments in automated 3D modeling.