What is scan-to-BIM?
Scan-to-BIM is the process of converting raw 3D point cloud data — captured by terrestrial laser scanners (TLS), mobile mapping systems, or aerial LiDAR — into structured Building Information Models that can be used for design, renovation, facility management, or construction coordination.
The output is a parametric 3D model (typically in Autodesk Revit, ArchiCAD, or IFC format) where walls are walls, pipes are pipes, and floors have elevations — not just a visual representation, but a semantically rich database of building components.
Why point clouds, not manual survey?
Traditional building documentation relies on tape measures, total stations, and manual sketches. For complex or existing structures, this approach is:
- Slow: A 50,000 SF facility can take weeks to manually document
- Incomplete: Concealed conditions, ceiling voids, and MEP routing are missed
- Error-prone: Human transcription errors compound across drawing sets
Terrestrial laser scanning captures millions of 3D measurements per second with millimeter-level accuracy. Field time varies significantly depending on scan density, access constraints, and the number of interior rooms, but scanning is substantially faster than manual documentation for complex facilities.
The seven-stage workflow
Stage 1: Project planning and control network
Before scanning begins, establish:
- Scope of work: LOD (Level of Development) requirements per the BIM Forum LOD Specification. LOD 200 (approximate geometry) is faster and cheaper; LOD 350 (precise geometry, connections, and supports) requires significantly more modeling time.
- Control network: Survey-grade control points (minimum 3, ideally 5+) established by total station and tied to the project coordinate system. These serve as ground truth for registration accuracy validation.
- Scan plan: Station locations planned to minimize occlusions and ensure overlap between adjacent scans (>30% overlap recommended for reliable registration).
Stage 2: Data capture
Terrestrial laser scanning (TLS): Instruments like the Leica RTC360, FARO Focus, or Trimble X7 capture 360° scans from tripod-mounted positions. Each scan station produces 50–300 million points depending on resolution settings. Typical capture rate: 15–40 stations per day for interior spaces.
Mobile mapping: Systems like NavVis VLX or GeoSLAM ZEB carry sensors through spaces continuously, trading some accuracy (typically ±15mm vs ±3mm for TLS) for dramatically faster capture. Ideal for large warehouses, corridors, and spaces where many TLS stations would be required.
Aerial LiDAR or drone photogrammetry: For roofs, facades, and site context. Drone-mounted LiDAR (DJI Zenmuse L2, Riegl miniVUX) or photogrammetric reconstruction from overlapping imagery.
Stage 3: Registration and georeferencing
Individual scan stations must be aligned (registered) into a unified coordinate system.
Target-based registration: Spherical or checkerboard targets placed in overlapping scan fields provide discrete correspondence points. Registration accuracy is reported as RMS error — typically < 3mm for well-planned projects.
Cloud-to-cloud registration: Algorithms (ICP — Iterative Closest Point, or feature-based methods) align overlapping point clouds without physical targets. Faster in the field but requires sufficient geometric features for reliable convergence.
Georeferencing: The registered point cloud is transformed into the project coordinate system using the control points established in Stage 1. This ensures the model aligns with survey drawings, site plans, and other project deliverables.
Software commonly used: Leica Cyclone REGISTER 360, FARO SCENE, Trimble RealWorks, or open-source CloudCompare.
Stage 4: Point cloud processing and classification
Raw point clouds contain noise, redundant data, and points from temporary objects (people, vehicles, furniture). Processing steps:
- Noise removal: Statistical outlier removal (SOR) filters eliminate isolated points caused by multipath, reflective surfaces, or sensor noise.
- Decimation (optional): Reduce point density for downstream software performance while preserving geometric detail at edges and transitions.
- Classification: Assign semantic labels to point cloud segments:
- Ground / floor
- Walls
- Ceiling / roof
- Structural (columns, beams)
- MEP (pipes, ducts, conduit)
- Clutter (furniture, temporary objects)
Classification can be manual (using cross-sections and intensity values), semi-automated (using geometric primitives and machine learning), or fully automated for simple geometries.
Stage 5: BIM modeling
This is the most labor-intensive stage. Modelers trace geometry from the classified point cloud into parametric BIM elements.
Modeling approach by system:
| System | Typical method |
|---|---|
| Walls | Fit planes to wall-classified points; set bottom/top constraints |
| Floors/ceilings | Fit horizontal planes; model as floor slabs with correct elevations |
| Structural | Identify column/beam cross-sections; model with steel or concrete families |
| MEP | Trace pipe/duct centerlines; size from point cloud cross-sections |
| Doors/windows | Identify openings in wall planes; place families with dimensions from cloud |
LOD considerations:
- LOD 200: Approximate geometry, ±25mm. Walls modeled as generic elements without detailed connections.
- LOD 300: Accurate geometry, ±15mm. Specific types assigned (CMU, metal stud, etc.).
- LOD 350: Includes connections, supports, and coordination-critical geometry.
Software: Autodesk Revit (dominant in AEC), Graphisoft ArchiCAD, Tekla Structures (steel/concrete), or specialized tools like ClearEdge3D Verity for automated pipe/duct modeling.
Stage 6: Quality assurance
QA verifies that the BIM model faithfully represents the point cloud data within specified tolerances.
Deviation analysis: Overlay the model surfaces on the point cloud and compute point-to-surface distances. Visualize as a heat map — blue (within tolerance) to red (exceeds tolerance). Tools: Verity by ClearEdge3D, CloudCompare (open source), or Revit-integrated plugins.
Acceptance criteria are defined per the USIBD Level of Accuracy (LOA) Specification, which is separate from the BIM Forum LOD Specification. LOA levels define positional accuracy:
- LOA 20: Points accurate to within 15mm (typical for existing conditions documentation)
- LOA 30: Points accurate to within 5mm (high-accuracy as-built)
- LOA 40: Points accurate to within 2mm (precision industrial)
Project-specific QA thresholds should be agreed in the scope of work. A common standard is 95% of modeled surfaces within the project’s stated LOA tolerance.
Completeness check: Verify all specified systems are modeled. Cross-reference against the scope matrix to ensure no areas or systems were missed.
Stage 7: Deliverables and handoff
Final deliverables typically include:
- Revit model (.rvt) or IFC export with classified elements
- Registered point cloud in E57 or RCP format for reference
- QA report with deviation analysis heat maps and statistics
- Coordinate system documentation (datum, projection, control point coordinates)
- Model element matrix documenting LOD achieved per system per area
File formats and interoperability
| Format | Use case |
|---|---|
| E57 | Industry-standard point cloud exchange (ASTM E2807) |
| LAS/LAZ | Aerial LiDAR standard (compressed with LAZ) |
| RCP/RCS | Autodesk point cloud format (indexed for Revit) |
| RVT | Revit native model |
| IFC | Open BIM exchange (buildingSMART) |
| FBX/OBJ | Visualization and rendering |
Common challenges
Occlusions: Areas not visible to any scanner (behind equipment, inside ceiling cavities) create gaps in the point cloud. Document these as “not surveyed” in deliverables — do not guess hidden geometry.
Reflective and transparent surfaces: Glass, mirrors, polished metal, and water cause false points or missing data. Supplemental manual measurements may be required.
Modeling ambiguity: Point clouds show surfaces, not internal composition. A wall might be CMU, metal stud with gypsum, or cast concrete — without destructive investigation, the internal composition is assumed based on building type and era.
Scale: A 200-station scan of a hospital can produce 10+ billion points. Processing and modeling requires substantial computing resources (64GB+ RAM, NVMe storage, capable GPU for visualization).
When to use scan-to-BIM
The investment in scanning and modeling is justified when:
- Existing drawings are unavailable, outdated, or unreliable
- Renovation or retrofit requires coordination with existing conditions
- Facility management needs an accurate digital twin
- Clash detection between new design and existing structure is critical
- The cost of field surprises during construction exceeds the cost of scanning
References
- BIM Forum LOD Specification (2024) — Defines Level of Development for model element geometry and data, not positional accuracy.
- USIBD Level of Accuracy Specification v3.0 — Defines positional accuracy levels (LOA 10–50) for as-built documentation.
- ASTM E57 Standard (E2807) — Standard specification for 3D imaging data exchange format.
- Leica Geosystems RTC360 Specifications — Range accuracy ±1.0mm + 10ppm at stated range.
- ISO 19650 — International standard for information management using BIM.