Conservation shophouses are the hardest kind of small building to document well. The floor plates are narrow, the stairs are steep and dark, the facades carry the ornament that makes the building worth conserving, and the interiors have usually been altered several times since the original drawings, if those drawings exist at all. A conservation practice preparing documentation for a Tanjong Pagar shophouse asked us to capture the whole building, inside and out, as one coherent dataset.
Why handheld SLAM for a shophouse?
A tripod scanner in a shophouse means dozens of setups: every landing, every half-turn of the stair, every room on every storey. We captured this one with a Lixel K2, a handheld SLAM scanner, so the operator simply walked the building while it recorded 200,000 points per second with real-time colourised preview. At roughly 1.2 kg it is light enough to carry up three storeys of steep timber stairs without a second thought, and SLAM does not need GPS, which matters in a deep, shaded terrace interior where satellite signal never reaches.
The trade-off is worth stating plainly, because it shapes when this method is the right one. A terrestrial tripod scanner delivers higher per-station accuracy, but every setup costs time and every pair of setups needs registration, and in a building made almost entirely of small, awkward spaces the setup count explodes. Handheld SLAM inverts the equation: continuous capture at walking pace, one trajectory instead of dozens of stations, and accuracy that is more than sufficient for conservation documentation, provided technique controls drift. For a shophouse, the geometry of the building itself makes that a straightforward choice.
The capture plan
We walked the route before scanning anything. The plan that worked:
- Start outside on the five-foot way with RTK fixed, so the scan opens with georeferenced coordinates and the facade ornament is captured at close range.
- Move through the entrance and ground floor in the same walk, tying the exterior and interior together in one trajectory rather than stitching separate scans later.
- Take the stairwells slowly, with deliberate overlap at each landing: tight vertical circulation is where SLAM drift is most likely, and steady pace prevents it.
- Close loops wherever possible, returning through spaces already scanned so the software can pull the trajectory back into alignment.
- Capture the rear court and party walls in a second, shorter walk, again starting with RTK outdoors.
The stairwell point is the one that repays the most attention. In a narrow vertical shaft the scanner sees very little geometry at any moment, and what it sees is repetitive: tread after tread, baluster after baluster. That is exactly the situation in which a SLAM trajectory can slip. Slowing down, pausing briefly at each landing so the sensor gathers a rich view of the surrounding rooms, and overlapping generously with the floor below gives the registration enough distinctive geometry to hold the trajectory together. Loop closure does the rest: every time the walk re-enters a space already in the scan, accumulated drift gets corrected against the earlier pass.
With roughly 90 minutes of capture per battery, each storey and the exterior fitted comfortably into separate scans with battery swaps between them.
Mixed indoor-outdoor capture in one coordinate frame
The five-foot way is the awkward zone: covered, so RTK is intermittent, but continuous with the street. Because the K2’s RTK is built in, the scans that began in open sky carried absolute coordinates into the covered walkway and the interior, and the whole building landed in one georeferenced frame at around 3 cm absolute accuracy, with roughly 1 cm relative accuracy in the details that conservation work cares about: mouldings, air vents, pintu pagar hinges, stair balustrades.
This is the quiet advantage of georeferencing a heritage record even when no drawing requires it. A purely relative point cloud is internally measurable but floats in its own local coordinates; a georeferenced one can be overlaid on cadastral data, on a neighbouring capture, or on a scan made years later, without any manual alignment. For a building whose documentation may outlive the practice that commissioned it, absolute coordinates are what keep the dataset useful to whoever opens it next.
Processing and the deliverables
We processed the scans in LixelStudio, merging the walks into a single registered point cloud and exporting a mesh alongside it. The routine: import each raw walk, run the SLAM optimisation so loop closures and landing overlaps tighten the trajectories, let the shared RTK frame place the walks relative to one another, then merge, clean the stray points that passers-by on the five-foot way contributed, and colourise. What went into the conservation record:
- A full-building colourised point cloud in LAS and E57, georeferenced via RTK.
- A mesh of the facade and key interior elements for elevation drawings and condition annotation.
- Sections cut straight from the cloud wherever the surveyor-drawn record was out of date.
The format choices are practical rather than ideological. LAS is the archival standard for point cloud data and opens in effectively every downstream tool; E57 is the vendor-neutral exchange copy for consultants on other software; and an RCP export from the same processed data drops straight into Revit or AutoCAD. The practice owns every file we handed over. The dataset is the permanent record of the building as found.
What made the difference
Starting every scan in open sky with RTK fixed before heading indoors. It costs a minute per scan and it is what turns a good relative point cloud into a georeferenced record another consultant can build on years later. The second was pace on the stairs: the whole job hinged on a few minutes of deliberate slowness in the two tightest parts of the building.
Documenting a conservation building?
The same method suits any conservation property in Singapore: shophouses, black-and-whites, places of worship, anything with tight circulation and ornament worth recording precisely. See how it fits heritage and urban planning work, read what 3D laser scanning covers, or send the address and what the record is for. We reply with scope and a quote within one business day.
