Researchers have introduced Rohbau3D, the first comprehensive 3D point cloud dataset that accurately captures indoor construction environments across multiple real-world sites.
Study: Rohbau3D: A Shell Construction Site 3D Point Cloud Dataset. Image Credit: Anatoliy Cherkas/Shutterstock.com
Published in Scientific Data, Rohbau3D features 504 high-resolution LiDAR scans collected from 14 high-rise construction sites in various stages of shell construction or renovation. The dataset offers a detailed, realistic foundation for advancing research in construction automation, digital modeling, and indoor scene understanding.
Background
In construction, detailed 3D models are typically developed only for large, high-budget projects where Building Information Modeling (BIM) justifies the cost. The vast majority of small and medium-sized projects, which make up most of the built environment, still rely on traditional 2D plans.
Renovation and repurposing projects face an even greater challenge: they often begin with no digital record of the existing structure. To improve efficiency and accuracy in these contexts, the industry needs reliable geometric data—and the tools to process it effectively.
While deep learning models are now capable of handling large 3D datasets, most existing resources focus on autonomous driving, infrastructure, or mixed-use interiors. Datasets specific to active construction environments are still rare. Rohbau3D fills this gap with high-fidelity 3D point clouds from real indoor construction scenarios.
How the Data Was Collected
The scans were collected over a one-year period in and around Munich, Germany, with four partners involved in site selection: a large construction firm, two smaller companies, and the Munich State Building Authority. This collaboration helped ensure the dataset reflects a realistic and varied cross-section of the industry.
Data was captured using a terrestrial laser scanner, capable of recording up to 1 million points per second, and paired with high dynamic range (HDR) spherical imagery to generate colorized point clouds. Mounted on a tripod for stability, the scanner was positioned by a trained civil engineer, who adapted placement to each site’s layout in order to minimize the number of scans while maximizing coverage.
Preprocessing was kept minimal to preserve the raw data’s integrity. A Euclidean distance filter removed points more than 25 meters from the scan origin to reduce noise caused by reflective surfaces and sensor anomalies. The positional accuracy of the scans was documented at 2 mm at 10 meters, 3 mm at 25 meters, and included an additional uncertainty of 0.1 mm per meter beyond that range.
What the Dataset Contains
Rohbau3D is a medium-scale collection of static indoor scans gathered from a broad spectrum of real-world construction sites. Each scan includes detailed 3D spatial coordinates, RGB color and surface reflectance data, reconstructed surface normal vectors, panoramic 2D images, and accompanying metadata. The dataset also comes with basic tools to help manage and interpret the data volume.
The 504 scans cover 14 distinct building environments, including six mid-rise residential buildings—four of them newly built, and two undergoing shell-preserving renovations. Beyond residential sites, the dataset features three large-scale school buildings with subdivided classrooms, a multi-story office building, an underground parking structure, and a vaulted brick cellar in a historical building.
To ensure clarity and expandability, the dataset was organized hierarchically. Each construction site was stored in a folder labeled with the prefix site_, containing nested folders for each individual scan (scan_). Within each scan, point cloud coordinates, annotation data (like color or surface normals), and panoramic images are saved as separate files.
Validating the Data
The researchers conducted a thorough sanity check to verify the completeness and consistency of each scene. While there were no ground-truth surface normals available for quantitative comparison, the predicted normals were assessed qualitatively for visual coherence, geometric accuracy, and structural consistency.
Some minor inconsistencies were observed in localized areas, but overall, the dataset demonstrates strong integrity and is suitable for a wide range of applications, including indoor scene geometry analysis, semantic segmentation, and surface reconstruction. Its modular design also makes it easy to extend or adapt for specific research needs.
Conclusion
Rohbau3D isn’t just a technical resource—it’s a reflection of what’s really happening on construction sites today.
By capturing the complexity and variety of indoor spaces during the shell phase, it gives researchers and developers the kind of real-world data they’ve been missing. Whether you're working on scene reconstruction, automation tools, or smarter renovation workflows, this dataset offers a solid, realistic foundation to build on. And as the industry continues moving toward more data-driven approaches, resources like Rohbau3D will be key to bridging the gap between design and on-site reality.
Journal Reference
Rauch, L., & Braml, T. (2025). Rohbau3D: A Shell Construction Site 3D Point Cloud Dataset. Scientific Data, 12(1). DOI: 10.1038/s41597-025-05827-7. https://www.nature.com/articles/s41597-025-05827-7
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.