What terrestrial laser scanning and photogrammetry actually measure on a live process unit
Terrestrial laser scanning measures range and two angles directly, so one setup is already dimensionally correct. Photogrammetry solves shape from overlapping images and has no size until a calibrated scale bar or surveyed control is added. Scanning wins inside dark, congested pipe racks. Photogrammetry wins on stacks, roofs and elevated steel a tripod cannot reach.
The two methods fail under opposite conditions, which is why plants that own a scanner still hire a drone. A terrestrial scanner emits its own 1550 nm light and times the return, so it works in a blacked-out compressor house at midnight and reports a distance whether or not the surface photographs well. Its weakness is geometric: it sees only what the tripod can see, and congested piping throws shadows that only extra setups fill. Photogrammetry carries neither light source nor scale. It infers shape from parallax across overlapping images, so it needs texture, even illumination and a known length in the frame. Given those, a drone documents the top of a stack with no scaffold at all. Large unit surveys combine both, registered into one coordinate frame, and the split is decided by access and surface condition rather than by resolution.
Source: Leica Geosystems, Leica RTC360 3D Reality Capture Solution datasheet (872750en, 07.21); FARO Technologies, FARO Focus Laser Scanner tech sheet (SFDC_04MKT_476, rev. 09/09/21); Trimble, Trimble X7 3D Laser Scanning System datasheet (PN 022516-364A, 09/19); Pix4D support documentation on relative and absolute accuracy of drone mapping; ASTM E2807-11(2019); ASTM E3125-17; VDI/VDE 2634; ISO 10360-13.
| Decision factor | Terrestrial laser scanning | Photogrammetry | What settles it in a plant |
|---|---|---|---|
| What is measured directly | One slope distance and two angles per point, referenced to the instrument's own origin | Nothing in 3D — a camera records bearings; shape is solved by bundle adjustment | Scanning is true to size at a single setup; photogrammetry is not true to size until it is scaled |
| Published 3D point accuracy at 10 m | Leica RTC360 1.9 mm; FARO Focus S 2 mm; Trimble X7 2.4 mm | Pix4D states relative accuracy of 1–2 × GSD horizontally and 1–3 × GSD vertically | Scanner accuracy is fixed by the instrument; photogrammetric accuracy is fixed by standoff and lens at the moment of flight planning |
| Conditions printed under that figure | Leica: 89% albedo, 68% confidence per JCGM100:2008. FARO: one sigma, after warm-up. Trimble: one sigma, instrument levelled within ±5° | Pix4D: a "correctly reconstructed project"; absolute accuracy cannot exceed the accuracy of image geolocation or ground control | Restated at the 95% confidence engineering specs normally use, every scanner figure doubles |
| Dark, sooty or black surfaces | FARO publishes range 0.6–350 m at 90% reflectivity but 0.6–50 m at 2% black; 10 m range noise rises from 0.1 mm to 0.9 mm across that span | Feature contrast collapses on uniform dark cladding, so image matching degrades or fails outright | Weathered jacketing and coke-covered steel punish both methods; scanning degrades more gracefully |
| Light and time of day | Independent of ambient light — the 1550 nm source is its own illumination | Requires even illumination; hard shadow, dusk and mixed lighting break image sets | Night shutdown windows are scanner territory |
| Occlusion in congested pipe | Line of sight only. Field of view 360° × 300° (RTC360) and 360° × 282° (X7) — the missing wedge sits under the tripod | Same line-of-sight physics, but a UAV occupies viewpoints no tripod reaches | Access, not accuracy, drives setup count and therefore the schedule |
| Getting into one coordinate frame | Targets, cloud-to-cloud alignment, or on-board tracking (Leica VIS, Trimble Registration Assist with IMU) | Bundle adjustment tied to ground control or calibrated scale bars; the effort moves into processing | Demand the registration or quality report either way, before accepting delivery |
| Deliverable, and what drives its cost | Indexed point cloud: E57 (ASTM E2807), LAS, RCP/RCS, PTX, POD. Cost tracks setup count and registration hours | Dense cloud, textured mesh and orthomosaic. Cost tracks image count, control effort and compute | Once CAD is requested, modelling hours dominate the cost of either method |
What each instrument actually measures
A terrestrial laser scanner is a polar measuring device. For every point it records one slope distance and two angles, then converts that triple into a Cartesian coordinate referenced to its own origin. Nothing external is required to make that coordinate dimensionally true. Put a scanner in front of a vessel, press the button, and the resulting cloud is already in millimetres — the instrument head is the ruler. This single property separates it from every image-based method, and it is the reason scanning became the default for industrial as-built work.
Photogrammetry measures nothing in three dimensions. A camera records a bearing from its perspective centre to a feature; a second camera in a different position records another bearing to the same feature; the intersection gives a point. A bundle adjustment solves all camera stations, all orientations and all object points at once. What emerges is geometrically correct in shape and completely undefined in size. The reconstruction of a 40-metre column and a 40-millimetre bolt are the same mathematics until something of known length enters the scene.
That difference drives everything downstream. Scale for photogrammetry comes from a calibrated bar laid in the scene, from surveyed control points, or from a laser scan used as reference geometry. Skip it and you receive a convincing model that measures wrong by a percentage no one detects by eye. Ask any photogrammetry vendor one question before all others: what provided the scale, and what is its stated uncertainty?
Reading a scanner datasheet without being misled
Three separate numbers appear on every scanner datasheet, and buyers routinely quote the wrong one. Range accuracy describes error along the beam. Angular accuracy describes error across it. Three-dimensional point accuracy combines the two, and it is the only figure that says anything meaningful about a measured point on steel. Leica publishes range accuracy of 1.0 mm + 10 ppm and angular accuracy of 18 arcseconds for the RTC360, then publishes 3D point accuracy of 1.9 mm at 10 m, 2.9 mm at 20 m and 5.3 mm at 40 m.
The growth with distance is the angular term opening into an arc. FARO publishes a ranging error of ±1 mm for the Focus S, defined in its own tech sheet as a systematic measurement error at around 10 m and 25 m, an angular accuracy of 19 arcseconds, and 3D point accuracy of 2 mm at 10 m and 3.5 mm at 25 m. Beyond 25 m the same sheet instructs the reader to add 0.1 mm of uncertainty for every additional metre — an explicit admission that the headline figure has a boundary.
Trimble publishes 2 mm range accuracy, 21 arcseconds angular accuracy, and 3D point accuracy of 2.4 mm at 10 m, 3.5 mm at 20 m and 6.0 mm at 40 m for the X7, valid when the instrument is levelled within ±5°. All three vendors state one sigma; Leica states 68% confidence per JCGM100:2008 explicitly. A specification written at the 95% confidence level that engineering documents normally use doubles every one of those numbers, which is the arithmetic sitting underneath most tolerance disputes on delivery.
What surface reflectivity does to range and to noise
FARO is unusually transparent about this and publishes range against reflectivity directly. For the Focus S the stated range is 0.6–350 m against a 90% reflectivity white target, 0.6–150 m against 10% dark grey, and 0.6–50 m against 2% black, all qualified as being for a Lambertian scatterer. The same instrument loses seven-eighths of its reach between a white calibration panel and a black surface. Nothing about the scanner changed; only the target did.
Noise follows the same curve. FARO's published range noise at 10 m for the Focus S Plus is 0.1 mm against 90% white and 0.9 mm against 2% black, and at 25 m the black figure reaches 1.6 mm. That is a factor approaching ten on identical hardware, driven purely by what the beam lands on. Incidence angle compounds it: a grazing shot along a pipe wall spreads the footprint and drags the computed point off the true surface.
Every manufacturer's headline number assumes a bright, square-on target. Leica states its RTC360 accuracy figures at 89% albedo. Trimble states range noise of under 3 mm at 60 m on 80% albedo at 1550 nm, and quotes its 0.6–80 m range on a matte surface at normal angle of incidence. A refinery offers none of those conditions. This matters most on exactly the surfaces you care about, because corroded, dark, weathered steel is also the steel whose geometry you want for thickness trending against real asset shape.
Where photogrammetry's accuracy comes from, and what caps it
Photogrammetric accuracy is governed by ground sample distance — the real-world size of one pixel on the object. Pix4D states that a correctly reconstructed project delivers relative accuracy of one to two times GSD horizontally and one to three times GSD vertically. GSD is set by sensor, focal length and standoff, which means the accuracy of the finished product is decided during flight planning, before a single image is captured. Flying higher to save time trades away millimetres in a way no processing recovers.
Absolute accuracy carries a separate and harder ceiling. Pix4D states that absolute accuracy cannot exceed the accuracy of the image geolocation or of the ground control points. A UAV using a standard GNSS receiver without control produces a model that is internally excellent and absolutely wrong by metres. Add surveyed control or an RTK/PPK platform and absolute accuracy rises to that of the control, never beyond the project's own relative accuracy.
Close-range industrial photogrammetry — the tripod-and-targets variety used for vessel deformation and large-part metrology — operates in a different regime and is evaluated under VDI/VDE 2634, with ISO 10360-13 now covering optical 3D systems internationally. Performance there is expressed as a maximum permissible length measurement error between targeted points across the measuring volume, established using reference scale bars. When a vendor quotes photogrammetric accuracy, establish which regime they mean; the two differ by orders of magnitude.
Occlusion decides more scopes than accuracy does
A scanner measures what its tripod can see and nothing else. In a congested pipe rack, every large-bore line casts a geometric shadow across everything behind it, and the only remedy is another setup from another angle. Setup count is therefore driven by congestion, not by the size of the area. A clear tank farm needs few stations across a wide footprint; a compressor house of the same footprint needs many, and the difference lands in both the schedule and the registration effort.
Field of view sets the residual blind spot. Leica publishes 360° × 300° for the RTC360 and Trimble 360° × 282° for the X7, which leaves an unscanned cone directly beneath the instrument. Combined with line-of-sight physics, this is why undersides of nozzles, tops of columns, the back faces of exchanger bundles and anything above a solid deck stay empty however many ground setups you buy.
This is precisely where photogrammetry stops being a competitor and becomes a complement. A UAV occupies positions that no tripod, ladder or lift reaches, and it does so without a scaffold permit. Combining aerial coverage of elevated steel with terrestrial coverage at grade, registered into one frame, produces the complete geometry that makes scaffolding and access planning before a turnaround worth doing at all.
Registration effort is the line item buyers forget
Individual scans are useless until they share a coordinate frame. Two classical routes exist: physical targets — spheres, checkerboards, surveyed control — placed so that each pair of setups shares enough of them, or cloud-to-cloud alignment that matches overlapping geometry directly. Targets cost field time and discipline; cloud-to-cloud costs overlap, meaning more setups and more data. Neither is free, and neither appears in the line of a quote that says only "scanning, per day".
Current instruments reduce but do not remove the work. Leica's RTC360 tracks scanner movement between setups with a video-enhanced inertial system to pre-align in the field. Trimble's X7 pairs an IMU-based Registration Assist with automatic refinement and produces a registration report carrying project and station average error, overlap and consistency results. Leica's Cyclone REGISTER 360 reports Bundle Error, defined as the average of Cloud-to-Cloud and Target errors, alongside Overlap and Strength metrics per link. Insist on that report as a contract deliverable.
Photogrammetry's equivalent cost sits in processing rather than in the field. More images, higher overlap and larger sensors all improve the bundle adjustment and all extend compute time. The failure mode is different too: a scanner registration that fails reports a large error, while a photogrammetric reconstruction across weakly textured surfaces can converge confidently on a subtly deformed answer. Verification, not vendor confidence, is what separates the two.
The deliverable each method naturally produces
Scanning produces an indexed point cloud. The vendor-neutral exchange format is E57, standardised as ASTM E2807, Standard Specification for 3D Imaging Data Exchange, Version 1.0 — a combined binary and XML container holding 3D points, their colour and intensity attributes, and 2D imagery captured by the same system. LAS, maintained by the American Society for Photogrammetry and Remote Sensing, dominates airborne and mobile LiDAR. Trimble's X7 exports TDX, TZF, E57, PTX, RCP, LAS and POD, which is a fair snapshot of what the industry actually moves between systems.
Photogrammetry produces a dense point cloud too, plus a textured mesh and an orthomosaic. The textured mesh is the format's great advantage in a boardroom and its great trap in engineering: a photo-textured surface looks more finished than any point cloud while carrying interpolated geometry between the points that were genuinely measured. Meshing invents surface. On smooth, low-texture steel it invents more of it.
Neither output is a CAD model, and neither knows that a flange is a flange. Everything above this line is measurement; everything past it is interpretation, performed by a person, priced in office hours. That boundary is the single most misunderstood point in a scanning enquiry, and it is worth reading what each rung of the deliverable ladder actually contains before writing a scope.
Choosing per asset rather than per site
The decision resolves cleanly once framed per asset. Interior, congested, dark, or requiring millimetre tolerance: terrestrial scanning. Elevated, unreachable, spread over a wide area, or needing centimetre tolerance across a large footprint: photogrammetry from a UAV. Assets that are both — a column with a congested base and an inaccessible top — take both methods registered together, which is standard practice on a full unit survey and should be priced as such from the outset.
Write four things into the enquiry and the quotes become comparable: the tolerance required and the confidence level it is stated at, the coverage expected including which elevations must be complete, the deliverable format and rung, and the coordinate frame the data must land in. Vendors who cannot answer against those four are quoting field days, not results.
The scan itself is a snapshot, and a snapshot ages the moment a field change is made. It becomes a digital twin only when it is maintained and joined to live inspection data — which is a different commitment from a survey, and one worth understanding against the tools plants already own. Atlantis runs both capture methods and the data layer that follows them, sized to the asset rather than the brochure. Affordable. Accessible. Fully customizable. Request a quote or a demo with your asset list and tolerance requirement, and the method split gets decided on evidence.
Does a laser scanner need ambient light to work?
No. Terrestrial scanners emit their own light. Leica and FARO both use a 1550 nm invisible Class 1 laser, and Trimble's X7 uses the same wavelength. The instrument measures time of flight on its own returned pulse, so a blacked-out vessel interior at 03:00 scans identically to noon. Photogrammetry has no source of its own and stops working when the light does.
How much range does a black or sooty surface cost you?
A great deal. FARO's published range for the Focus S is 0.6–350 m against a 90% reflectivity white target, 0.6–150 m against 10% dark grey, and 0.6–50 m against 2% black, all for a Lambertian scatterer. Range noise at 10 m rises from 0.1 mm on white to 0.9 mm on black. Sooty insulation jacketing sits at the punishing end of that scale.
Can photogrammetry replace laser scanning for as-built piping?
Not for congested process piping. Photogrammetry needs texture and multiple unobstructed viewpoints on every surface, and a pipe rack denies both — bare pipe is smooth, uniform and self-occluding. Photogrammetry earns its place on the assets a tripod cannot reach: stack and flare tips, tank roofs, elevated structural steel, and large-area site context captured from a UAV.
What provides the scale in an industrial photogrammetry job?
A calibrated scale bar placed in the scene, surveyed control points, or a laser scan used as the reference geometry. Close-range industrial photogrammetry systems are evaluated under VDI/VDE 2634 and ISO 10360-13 as a length measurement error across the measuring volume, established with reference scale bars. Ask for the scale reference and its stated uncertainty before you look at any other spec.
Why do two scanners with similar range accuracy give different point accuracy?
Because the angular term dominates at distance. Leica publishes 18 arcseconds angular accuracy, FARO 19 arcseconds and Trimble 21 arcseconds. That angle opens into an arc as range grows, which is why Leica's 3D point accuracy moves from 1.9 mm at 10 m to 5.3 mm at 40 m, and Trimble's from 2.4 mm to 6.0 mm across the same span.
Which method copes with a live unit while people and vehicles move?
Scanning, with the right feature enabled. The RTC360 offers a double scan mode for automatic removal of moving objects, comparing two passes and discarding what changed. Photogrammetry requires the scene to hold still across an entire image set, so operators walking through a flight line contaminate the reconstruction rather than being cleanly rejected from it.