Most SLAM captures live happily in their own local coordinates. This one could not: the team needed a point cloud of a dense CBD site that would sit correctly against cadastral data and an existing site model, which means absolute coordinates, which means RTK. We captured it with a Lixel K2, chosen specifically because its RTK receiver is built in rather than bolted on, and planned the capture around one question: where on this site will RTK actually work?
Why RTK, and where PPK fits
A quick orientation. RTK, real-time kinematic positioning, corrects the scanner’s GNSS position live against a correction stream, so the trajectory acquires absolute coordinates during the walk itself. The alternative, post-processed kinematic (PPK) workflows, records raw GNSS observations and resolves them against base-station data after the fact. PPK can rescue a capture where a correction link is unreliable, but RTK has a decisive advantage on a difficult site: the operator can see, while walking, whether a fix is held or lost, and adapt the route on the spot. On a site whose entire risk was sky visibility, that live feedback was the point, and it is why the built-in receiver on the Lixel K2 mattered more than any spec-sheet number.
RTK behaviour between towers
Dense urban Singapore is close to a worst case for satellite positioning. Towers block half the sky, glass facades bounce signals into multipath, and covered linkways cut reception entirely. The pattern we found matches what any operator should expect:
- Open plazas and the road reserve gave quick, stable RTK fixes.
- Street canyons between towers were intermittent: fixes came and went, and the K2 rode through the gaps on SLAM, re-anchoring whenever the fix returned.
- Under overhangs, void decks, and indoors there is no RTK at all, and the trajectory is pure SLAM.
The practical rule that followed: start every walk in the most open sky available, hold still until the fix is solid, and route the walk so it revisits open-sky ground periodically instead of spending the whole battery in the canyons. Each re-anchor acts like a loop closure against the world itself: the SLAM trajectory that drifted slightly through a canyon gets pulled back onto absolute coordinates the moment the sky opens, and the correction propagates back through the intervening geometry in processing.
Control points before trusting anything
Built-in RTK is not a substitute for checking. Before capture we marked a handful of well-defined points across the site: kerb corners, drainage grates, paint marks on open ground, several with known coordinates from the project’s existing survey. Every processed walk was compared against them. Where a scan agreed with control within the expected envelope we accepted it; one walk that had spent too long between towers disagreed noticeably, and we re-captured it with a better route rather than forcing it to fit.
That last decision is the methodological heart of the case study. Registration software will happily warp a bad trajectory onto control points and report success; the errors do not disappear, they just hide between the constraints. Re-walking a suspect scan with a route that touched open sky more often cost an hour; a silently distorted cloud would have cost far more downstream, when measurements taken between control points stopped agreeing with reality. Independent check points, ones the alignment is never fitted to, are what let us say a cloud is verified rather than merely consistent.
Merging multiple walks into one cloud
The site was too large for a single battery, so the capture became five overlapping walks, each roughly a scene of up to 90 minutes, each starting and ending in open sky. Overlap zones were planned deliberately: shared corridors of geometry a few metres wide where adjacent walks saw the same facades and street furniture. In processing, the shared RTK frame did most of the alignment, the overlap geometry refined it, and the merged cloud was then re-checked against the control points as a whole before anyone downstream touched it.
Processing followed the standard sequence: each walk imported and SLAM-optimised individually, georeferencing applied from the RTK data, then the walks merged, cleaned of transient points from traffic and pedestrians, and exported. Because the downstream users ranged across GIS and design tools, we delivered the merged cloud in LAS for the primary dataset with E57 as the vendor-neutral exchange copy, either of which drops into a handheld 3D mapping or overlay workflow without translation losses.
What accuracy should you expect: absolute vs relative?
The two numbers do different jobs, and conflating them causes most georeferencing disappointment. The K2’s relative accuracy, around 1 cm within a local neighbourhood of the scan, is what makes measurements inside the cloud trustworthy: a doorway width, a kerb height, the gap between two structures. Absolute accuracy, around 3 cm RMSE with a solid RTK fix, is what places the whole cloud on the map, and it degrades where the sky does. On this site the open areas verified comfortably against control, while deep canyon stretches were treated as a few centimetres looser. For overlaying against cadastral and model data that was exactly fit for purpose; had the specification demanded survey-grade certification, that would have been a different scope with a surveyor in it, which is why we ask what the coordinates are for before quoting.
What was handed over
Five raw scans, one merged and control-checked point cloud in absolute coordinates, the control check report showing absolute against relative, and a capture log noting where RTK held and where it did not, which is the document that will save the next capture on this site half a day of guesswork.
Need a site in absolute coordinates?
The recipe transfers to any dense site: mark check points before scanning, start and end every walk in open sky, plan overlap corridors between walks, and verify the merged cloud against control it was never fitted to. See what land survey and georeferenced capture covers, or the full capture-to-deliverable workflows. Send the site, the coordinate system and the accuracy you need to hold, the last one matters most. We reply with scope and a quote within one business day.
