A BIM consultancy was tasked with modelling existing conditions for an older building ahead of a refurbishment. The record drawings were incomplete and decades of minor changes had gone undocumented. Before any modelling could start, they needed a reliable as-built reference layer.
They brought us in for the capture. We scanned and processed the building with a Lixel K2 and handed over reference; the modelling stayed in-house, where the BIM expertise and the client relationship lived. We model to BIM too when a client wants the whole chain, but on this job the split was the point.
Point cloud first, model second
We captured a registered point cloud of the building in a series of planned walks. Processed and exported as LAS, E57 and RCP, that cloud drops straight into the BIM environment as a measurable backdrop the modellers trace and verify against.
Keeping capture and modelling separate was deliberate. We controlled schedule and site access for the capture; the consultancy controlled model quality and LOD. Neither waited on the other.
The sequencing matters more than it first appears. A model traced from incomplete drawings inherits their errors and then adds interpretation on top; a model traced against a point cloud starts from measured reality. Walls that drift off-grid, floors that are not level, openings that were moved decades ago and never redrawn, all of it is simply present in the cloud, and every modelling decision can be checked against geometry rather than memory. The cloud is not the deliverable; it is the evidence the deliverable rests on.
Planning the capture walks
An older, larger building is captured as multiple walks rather than one heroic trajectory, and we planned the walks before arriving on site. Each walk is a closed loop where the floor plan allows, because loop closure, returning through space already scanned, is what corrects the drift every SLAM trajectory accumulates. Adjacent walks share overlap zones, stairs, lobbies, corridors seen by both, so the processing knits them into one registered cloud. Where the building met open sky, starting a walk outdoors with RTK fixed gave the dataset absolute coordinates that carry indoors; deep-plan interiors run on pure SLAM, which is exactly what the device is designed for. Where tighter verification is required we add control: well-defined points with known coordinates that the processed cloud is checked against rather than fitted to.
Capture and model, in lockstep
The two streams stayed coordinated through four checkpoints:
- Agree. Set tolerance, LOD expectations, and the formats the BIM environment needs before any capture.
- Capture. Walk the building, planning control and loop closures for larger structures.
- Export. Process and deliver LAS / E57 / RCP reference for Revit or AutoCAD.
- Model. The consultancy models against the reference, with us on hand for coverage questions.
The processing run is the same for every walk: import, SLAM optimisation so the planned loop closures pull the trajectory tight, registration of the walks into one cloud, then cleaning, cropping the site boundary and filtering the transient points that occupied buildings always contribute. The export step is dictated by the modelling environment: RCP for direct referencing in Revit and AutoCAD, LAS as the archival master, E57 when a consultant needs a vendor-neutral copy. Inside Revit the modellers worked floor by floor, cutting live sections through the cloud, tracing structure and envelope first, then partitions and openings, verifying each element against the points before committing it to the model.
The lesson worth passing on is about sequencing, not software: reviewing each processed walk in the same week is what keeps a multi-day capture honest. Checking the registered cloud for gaps while we are still mobilised means a missed plant room is a half-hour detour, not a second visit. The other habit worth keeping is a plain-text capture log, which walk covered what, where access failed, which areas were furnished or altered, because the modellers reading the cloud months later will ask exactly those questions.
Do you need a survey-grade scanner for scan to BIM?
It depends entirely on the tolerance the model must meet, which is why tolerance is checkpoint one. Device specifications describe ideal conditions. What reaches the model depends on capture technique, site conditions, control strategy, and processing. For tolerance-critical or compliance work we agree requirements with the modellers in advance and state them in the quote, so the right method and control are arranged. These are reference workflows, not certified survey deliverables. Where survey-grade control is required, plan for additional control and verification.
- Around 1 cm relative accuracy at device level
- Real-world results depend on technique and conditions
- Define tolerance and LOD before capture
- Reference workflow, not a certified survey
For existing-conditions modelling ahead of a refurbishment, the honest answer is that handheld SLAM reference is very often sufficient: the modelling decisions it informs, where a wall is, how a floor plate actually sets out, what was built in place of the drawing, operate at tolerances the K2 meets comfortably. The cases that need more are the ones that need certification or millimetre-class control, and those should be scoped that way from the start rather than discovered late.
One boundary stayed clean throughout: we captured, processed and delivered the reference; the consultancy built the model against it. Where a client wants the whole chain instead, we take it through to the Revit model at an agreed LOD.
Modelling against an older building?
The pattern travels well: agree tolerance and LOD first, plan closed-loop walks with overlap, and hand the modellers RCP reference they can trace in place. Read what Scan-to-BIM covers end to end, or see the same capture serving asset records on facilities and asset documentation. Send the building type, area and tolerance expectations and we will confirm scope and a quote within one business day.
