An R&D team at Austrian digitization platform Metaroom by Amrax recently captured a fully operational 60,000-square-metre German hospital using their iPhone LiDAR app. The entire project was completed within the Metaroom platform, producing a single structured parametric model rather than a conventional point cloud or mesh. The project offers a useful measure of how far consumer LiDAR has progressed for large-scale as-built documentation, while also illustrating where terrestrial laser scanning remains the more appropriate tool. This article examines what was delivered, how the workflow performed, and the trade-offs practitioners should consider. The broader significance, however, goes beyond the cost or speed of a single project: workflows like this can make digitization viable for buildings that previously could not justify scanning, while lowering the barrier to recapturing them often enough to keep their digital models current.
What Was Actually Captured
The headline number is 60,000 square meters (646.000 ft2), but the more telling figure is the room count. The team documented roughly 1,500 rooms of nearly every type a large hospital contains: offices, toilets, technical and plant rooms, a data center, restaurants, a rehabilitation and swimming pool area, and operating theatres. That mix matters, because it is the awkward, cluttered, reflective, and access-restricted spaces that usually expose the limits of a capture method.
The Numbers: People, Time, and Cost
The capture took six days on site, with each session running for about six hours. Three people scanned in parallel, and a single person handled four days of post-processing and export afterwards in the browser-based Metaroom Workspace. Start to finish, the project ran to roughly two working weeks. Capture does not require a trained surveyor; Metaroom is built to be used by non-specialists after about a day of training, which is part of why three people could cover a building this size at once.
It is worth setting the economics against the conventional alternative. Terrestrial LiDAR scanners cost between €20,000 and €100,000 per device, and the hardware is only the entry fee: converting the resulting point cloud into usable BIM geometry is specialist work that routinely costs several times more than the scanning itself, and for a large hospital the combined bill commonly approaches €500,000. By comparison, the Metaroom capture and model of this hospital came in at around €40,000. The interesting question is not whether the phone-based approach is cheaper, but whether what it produces is good enough to replace that workflow rather than sit alongside it.
Parametric, Not a Point Cloud, and Why That Distinction Matters
The constraint that makes the project interesting is that the hospital never closed. Nothing was taken offline for documentation. Clinical areas that are difficult to access at the best of times, including operating theatres, were captured while they remained in use. Anyone who has tried to schedule a terrestrial scanner around a live healthcare facility will recognize how much friction that removes. The work was commissioned in connection with a Siemens project, and the team behind the Metaroom platform describes the result as what it believes to be the largest parametric building model captured to date using a smartphone.
This is where the project departs from most mobile capture stories. The output is not a point cloud or a mesh. It is a structured parametric model in which every wall, door, window, and opening is represented as a discrete, measurable object with a defined position, height, and thickness. Rooms and floors are detected automatically as well, and the individual captures are assembled into a complete building structure rather than a loose collection of scanned spaces.
The difference is not cosmetic. A point cloud or mesh is accurate but carries no structured data; a wall is simply a surface where adjacent objects blend together. A parametric wall is something a BIM manager, MEP engineer, energy consultant, or facility manager can work with directly in a planning tool. Traditional workflows, including those built on professional hardware, deliver a point cloud that still has to be modeled into geometry by hand, a process that can take weeks. Here, the structured model is generated automatically and is available at the moment the scan is complete. In the current beta, it exports as IFC, 2D DXF, or a 2D PDF floor plan, with the platform supporting more than 40 formats in total as further export options are rolled out.
The Technology: RoomPlan, Semantic Segmentation, and Scan Merging
Metaroom uses Apple’s RoomPlan API as its scanning foundation and extends it with a proprietary stack. LiDAR depth data is fused with camera-based computer vision and processed through deep learning models for indoor semantic segmentation, which is what allows the system to identify and classify architectural elements automatically rather than handing back an undifferentiated surface.
Within a single scan, SLAM continuously tracks the device’s pose and builds a map of the space, registering all captured data into one consistent 3D model as the operator moves. Combining many independent scans into one coherent building is a separate problem, and Metaroom handles it with a purpose-built, easy-to-use toolset in the Workspace that lets the user merge the individual captures into a single large model without specialist knowledge. The platform takes care of the technical details so the final model meets the required structural and integrity standards. The hospital was the first large-scale test of that workflow, validating it in a live building rather than an empty shell.
Geometric and Classification Accuracy
Locally, geometric accuracy is typically within approximately 1% deviation. For example, if two walls in a room are five meters apart, the modeled distance would typically be within about five centimeters of the true dimension. At this scale, the accuracy is broadly consistent with USIBD LOA 10, where 95% of measurements fall within the applicable 5–15 cm accuracy range.
Globally, accuracy is governed primarily by SLAM drift, which is typically on the order of 1% of path length. Over a 20-meter scanning path, for example, accumulated error could be on the order of 20 centimeters. Effective data-collection practices—such as keeping scan extents reasonable—and corrections from loop closure help constrain the accumulation of SLAM drift.
The resulting model is generally consistent with LOD 200, representing approximate geometry, locally approaching LOD 300 in some cases.
As noted previously, the parametric model classifies walls, doors, windows, furniture, and other room-defining features. These classifications are reliable for the primary elements needed to characterize a room.
Metaroom is not intended to compete directly with the geometric accuracy, classification detail, or level of detail achievable through traditional terrestrial laser scanning workflows with manual modeling and classification. Instead, it provides a faster and more cost-effective alternative for applications where millimeter-level accuracy and high point density are not needed. Where that level of performance is sufficient, the economics of reality capture change significantly—making more projects practical to scan and allowing existing conditions to be updated more frequently.
Metaroom is planning a case study that will provide a more formal assessment of these performance metrics, allowing its results to be compared more directly with established reality capture workflows.
What It Means for Building Professionals
Large buildings are precisely where conventional digitization workflows break down: floor plans are outdated or missing, site visits get repeated, and redrawing consumes weeks. A method that lets a small team capture a 60,000-square-metre facility in days and hand off a model that goes straight into BIM, MEP, energy consulting, lighting design, and facility management is addressing a real and expensive bottleneck. The hospital project demonstrates that smartphone capture is an efficient and effective option for projects that do not require high accuracy and detail.
The larger opportunity is not simply reducing the cost of an individual scan—it is expanding when reality capture makes economic sense. Lower-cost capture can bring projects into reach that previously could not justify a terrestrial laser scanning workflow, allowing more buildings to benefit from accurate existing-conditions data.
It also changes how often that data can be refreshed. Building models inevitably diverge from reality as renovations, equipment changes, and other modifications accumulate. When recapturing a facility is expensive, those discrepancies are often allowed to persist. A faster, more affordable workflow makes periodic rescanning practical, helping BIM and facility-management models remain much closer to the buildings they are intended to represent.
Together, these changes broaden access to reality capture while shifting it from an occasional documentation exercise toward a more continuous source of reliable building information.
This does not prove that smartphone capture replaces terrestrial scanning everywhere. It suggests that for a large and growing class of as-built work, the question is no longer whether the phone is good enough, but which jobs it is now good enough for.
But the shift that interests me most is not any single capture. It is frequency. A traditional survey is expensive enough that it happens once, if at all, and the model starts going stale the moment a wall moves. When a scan is this cheap and fast, the model no longer has to be a one-off. It can be re-captured after each renovation, or on a routine facility-management cadence, and kept in step with the building it describes. A living, affordable-to-refresh model is something a high-cost survey structurally cannot offer. And the same economics reach further still: they make digitization viable for the buildings that were never documented in the first place, simply because terrestrial scanning was too slow or too expensive to justify. That, more than the 60,000 square metres, is what the hospital really demonstrates.
Key Takeaways
- A live, 60,000 m² hospital, around 1,500 rooms across every functional type, was captured on iPhone Pro devices while the building stayed fully operational, including operating theatres in use, with the whole operation run on the Metaroom platform.
- The workflow ran roughly two working weeks: six days of on-site scanning by three people in parallel, plus four days of post-processing and export by one person. Capture needs only about a day of training rather than a specialist surveyor.
- The deliverable is a structured parametric model: walls, doors, windows, openings, rooms and floors detected automatically and assembled into a complete building, not hand-modelled from a point cloud over weeks.
- The stack is Apple’s RoomPlan plus deep-learning semantic segmentation; As you move through a space, SLAM continuously registers each individual scan and aligns them into a single consistent model., and a Workspace toolset merges separate scans into a single building without specialist knowledge.
- Judge it on real numbers, not a headline figure: roughly 1% local deviation (centimeter-scale on room dimensions), global accuracy governed by about 1% SLAM drift. LOA10 locally and LOD 200 geometry. It is an as-built and planning tool, not a replacement for terrestrial scanning where tight tolerances or high LOD are non-negotiable.
- Against a traditional scan-to-BIM bill that can approach €500,000 for a building this size, the Metaroom capture and model came in at around €40,000. Beta exports to IFC, 2D DXF, and PDF, covering a wide range of software and use cases.
- The real opportunity is frequency: capture is cheap and fast enough to keep a model current after every renovation, and to digitize the buildings that slower, costlier methods left undocumented altogether.
Availability
Metaroom’s Enterprise Building Capture, the same feature built to complete the hospital project, is available as a public beta. The Metaroom Scan App runs on LiDAR-equipped iPhone Pro and iPad Pro devices via the Apple App Store, and Enterprise Building Capture is accessed through the browser-based Metaroom Workspace. Account registration and beta access, along with details of upcoming webinars, are available at amrax.ai.
Author Details
This post was contributed by Metaroom by Amrax and edited by Nathan Roe.
