Thickness decides how much filling a sandwich biscuit takes, how much dough a line wastes, and whether the pack seals. A laser line profile over the belt measures that thickness on every part, at full line speed, without touching anything. The geometry below sets your height resolution, the line-rate arithmetic keeps you in step with the belt, and a two-block calibration turns pixels into millimetres — all worked through for a 100 mm field of view, a 2K sensor and a 150 mm/s belt.
How does a laser line profile turn height into pixels?
A laser fan painted across the belt and a tilted camera watching it turn height into sideways shift: on the 100 mm, 2K layout below, 1 mm of height moves the imaged line about 10 pixels, and that ratio is your height resolution.
The laser module projects a fan — a thin plane of light standing vertical across the belt, perpendicular to travel. The camera sits tilted so it watches the line from the side. Where the belt is bare, the line lands at one image row; where a part lifts the surface, the line shifts across the image by height × tan θ, with θ the camera tilt from the belt normal. Stacking one profile per encoder tick builds the height map: columns across the part, rows along the travel.
Three hardware facts follow. Laser and camera mount on one rigid plate, because every micron of relative movement reads as height. The fan stays perpendicular to travel so each part is cut at the same station. And a profile only contains what the fan crosses: presence, width at the cut and height come from the profile stack, while full footprint and surface colour belong to a separate top-down station. The pipeline from raw profiles to a usable height image is what How to create a 2D image from laser triangulation walks through step by step.
What camera angle gives the height resolution you need?
Tilt the camera about 27 degrees from the belt normal — a 1:2 slope. A 100 mm field of view on a 2048-pixel sensor then reads 0.049 mm per pixel laterally and about 0.10 mm of height per pixel, which covers bakery thickness tolerances of a few tenths of a millimetre.
Worked through: 100 mm of field on 2048 pixels gives 100 / 2048 = 0.049 mm per pixel laterally. At θ = 27° (tan θ = 0.5) the height per pixel is 0.049 × 2 = 0.098 mm — call it 0.10 mm. Read the other way, 1 mm of height shifts the line 1 × 0.5 / 0.049 ≈ 10 pixels. A 10 mm height range therefore sweeps about 102 pixels of image travel, which is what your sensor span and focus latitude must cover.
| Camera tilt θ | tan θ | Height per pixel | 100 mm field on 2K | Fits |
|---|---|---|---|---|
| 14° | 0.25 | 4 × lateral | 0.20 mm | coarse presence checks, tall parts |
| 27° | 0.50 | 2 × lateral | 0.10 mm | bakery thickness, packaged food |
| 45° | 1.00 | 1 × lateral | 0.05 mm | rigid parts, fine steps |
Steeper buys finer height steps and costs image span: at 45° that same 10 mm range sweeps 204 pixels, and the lens must hold focus across the full oblique path. Shallower buys depth-range headroom and coarse numbers. For food thickness in the 0.1 mm class, 27° is the trade to take.
Centroid extraction finds the line centre to a fraction of a pixel — around a quarter pixel on clean matte surfaces, nearer half a pixel where laser speckle from rough crumb dominates. Treat interpolation as margin: plan the tolerance at pixel level, 0.10 mm here, and let the centroid improve repeatability from there. Laser line extraction noise: get a clean height line shows what the centroid is fighting.
How fast must the line rate be to keep up with the belt?
Set the profile spacing first, then divide: at 150 mm/s belt speed and 0.25 mm spacing you need 600 lines per second, which is 240 profiles over a 60 mm biscuit — and any line scan camera rated in tens of kilohertz passes that with a wide margin.
The arithmetic: 150 / 0.25 = 600 lines per second. A 60 mm biscuit spends 60 / 150 = 0.4 s in the fan, so 600 × 0.4 = 240 profiles cover it — one height sample every 0.25 mm along its length. Line scan cameras are rated in the tens of kilohertz, so the 600 lines/s budget closes with an order of magnitude to spare.
Blur comes from travel during exposure. Hold blur to one lateral pixel — 0.05 mm — and exposure must stay under 0.05 / 150 = 0.33 ms. A laser line concentrates its light onto a narrow strip, so a few hundred microseconds of exposure is normal and the blur budget closes with margin.
Drive the capture from a conveyor encoder scaled to one profile per 0.25 mm of travel, and belt speed stops mattering: spacing follows the belt instead of the clock, so every part gets the same sampling however the line surges. Encoder ticks per pixel for line scan: set the scale right works the tick scaling. The same blur-versus-speed budget, on plain 2D webs, is in camera fps and shutter speed for moving web inspection.
Do you buy an integrated profiler or build from a camera and laser?
Buy an integrated laser profiler for most conveyor height work: laser, receiving optics and calibration ship as one factory-aligned unit. Build from a line scan camera plus a machine-vision line laser only when you need an odd field of view or want to reuse a camera platform you standardize on.
A profiler is a factory-aligned pair — laser, receiver optics set at the triangulation angle, and the calibration model — in one housing that reads out millimetres. You mount it, wire trigger and encoder, and own the application rather than the optics. The build path trades that alignment for freedom: any field width, any pixel count, reuse of hardware you already run, at the price of choosing the lens, focusing it and owning the calibration yourself.
Both profilers in the cards are integrated laser profile sensors made for this mount-over-the-belt job. Compare their printed specs, then check stand-off and profile width against your belt layout in the 3D laser profile sensors range. Onboard data reduction is part of how profilers hold their rated rate, which 3D profiler profile rate: how onboard data reduction enables speed explains.
- Type
- Laser profiler
- Type
- Laser profiler
For the build path, a 2K line scan camera is the workhorse. The LS-LCM-0016 pairs 2K resolution with 7 μm pixels and a 58 kHz line rate on GigE, in an M42 mount for the fixed-focal lens this geometry needs; onboard TDI adds sensitivity at the short exposures this laser-and-speed budget sets. Add a machine-vision line laser whose fan covers the 100 mm field plus margin, a rigid common mounting plate, and an encoder. The line scan cameras range covers the sensor alternatives. If the task also needs missing-part or volume logic, Industrial 3D camera selection: stereo, ToF or structured light places laser profiling among the options.
- Lens mount
- M42
- Interface
- GigE
- Line length
- 2K
- Pixel size
- 7 μm
Whichever path you take, the calibration in the next section is yours to run.
How do you calibrate the height scale on the line?
Calibrate with two known heights: scan the bare belt for the zero plane, then scan a gauge block of known thickness for scale — on the 27-degree layout a 2.00 mm block shifts the line about 20 pixels, and your measured count is the millimetres-per-pixel.
- Set the focus and lock every adjustment at mid-height of the working range, then leave the plate alone.
- Scan the bare belt and record the line's image row: that row is your zero plane.
- Scan a 2.00 mm gauge block. Expect the line to move about 20 pixels; the measured count Δpx sets the scale as 2.00 / Δpx millimetres per pixel, about 0.098 here.
- Repeat with a second block — 5.00 mm — and confirm the scale agrees within about 1%, which catches tilt or lens non-linearity.
- After any bump, refocus or part swap, run the pair again. The geometry is fixed, so the scale holds; only the zero drifts.
Belt surfaces vary too. Subtract the belt's own line from each profile so height reads relative to the belt directly under the part — that cancels belt thickness flutter and roller dip, and it is the difference between a measurement and a number. Sheet of light laser triangulation: the calibration model explained derives the full model these two blocks are pinning down.
Where does laser line profiling fail on food or shiny product?
Matte, mid-brown surfaces profile cleanly; glazed, wet or near-black surfaces fight the laser. A shiny glaze mirrors the line into spikes and ghost reflections, dark cocoa swallows red light, and loose film splits the line into two surfaces — fix wavelength, exposure or angle before blaming software.
Matte baked dough is the friendly case: it scatters the line evenly and the centroid stays clean. Trouble starts with surfaces that mirror or absorb. A sugar glaze acts like a mirror: the line spikes, saturates, and part-to-belt reflections add a ghost line below the real one. Drop the exposure so the spike clips and can be rejected, angle the laser a few degrees off the camera plane so direct reflections leave the frame, and keep the belt matte.
Dark cocoa and currant-rich doughs absorb red light, and flour dust in the air scatters it. Wavelength is the fix: dark surfaces profile better toward blue or near-infrared, both of which mono sensors still see. Clear film wraps split the line into top and bottom surfaces, so either pick a wavelength the film passes or define the measurement on one surface. Dust settling on windows needs an air purge and a wipe-down schedule from day one.
Height belongs to the laser; length, width and surface colour belong to a top-down station with line scan lights. Asking a tilted profile camera for a plan view distorts both.
And if the real target is bulk depth or volume on a conveyor — a loading profile rather than part height — a laser versus ultrasonic conveyor measurement comparison lays out where point sensors and ultrasonics still win.
What do you check before ordering?
Close four numbers before you order: field of view, height resolution, height range and belt speed. For a 100 mm field, 0.1 mm-class resolution, a 10 mm height range and a 150 mm/s belt, the layout above closes at 600 lines per second with 0.33 ms exposure.
- Field of view — widest product plus 10%. A part at the belt edge must still cross the fan.
- Height resolution — tolerance divided by ten. A ±0.5 mm thickness tolerance needs the 0.05-0.10 mm class from the tilt table.
- Height range — tallest product plus 2× margin. Belt flutter, focus latitude and the tallest sandwich all live in that margin.
- Camera tilt — 27° for food thickness; steeper only for rigid parts with fine steps.
- Line rate — the 600 lines/s this layout needs is a floor, not a target; leave a factor of two.
- Exposure — the 0.33 ms blur budget here; a laser line meets it easily.
- Laser class — Class 2 over an open belt; guard and interlock anything stronger.
- Mounting and encoder — camera and laser on one plate, encoder on the drive, gauge blocks in the drawer.
Load the four numbers into the camera configurator and it narrows sensors to the field, rate and interface you computed — or send the belt layout through the contact page and we will check the geometry with you. The rest of the triangulation toolkit lives on the 3D topic page.
Questions
What laser class is safe over an open production line?
Class 2 — under 1 mW — is the default for an exposed belt: the eye's blink reflex covers accidental glances, though you still mount the head above reach and label it. Anything Class 3B needs fixed guarding, an interlocked enclosure and laser-safety oversight under IEC 60825-1. Treat stray specular reflections off shiny product as exposure paths too.
Can one camera measure length, width and thickness at once?
Not well in one pass: the 27-degree tilt that gives height sensitivity foreshortens the plan view, so length and width come out distorted. Standard practice is two stations — a top-down camera with LED lighting for footprint, then the laser profile for thickness — or an integrated profiler whose software reconstructs both from the profile stack plus the encoder record.
How often does the height calibration need repeating?
Re-zero against the bare belt at every shift start — one scan, and it catches stand-off drift from bracket temperature and belt tension. Repeat the full two-height calibration only after a mechanical event: a bumped bracket, a lens refocus, a camera or laser swap. Between events the geometry is fixed, so the millimetres-per-pixel scale holds; only the zero moves.
Does the red laser line read on dark chocolate or wholegrain surfaces?
Wholegrain and mid-brown baked surfaces scatter red light well and profile cleanly. Very dark cocoa or currant-rich dough absorbs 660 nm strongly: the line dims and the centroid gets noisy. The fixes in order of cost are longer exposure, a brighter line laser within the same class, then a switch to blue around 450 nm or near-infrared around 808 nm, both of which mono sensors still read.
Can height data estimate product weight?
Height times footprint times bulk density gives a usable mass estimate, typically within a few percent while the recipe stays stable — enough to steer the depositor before a checkweigher rejects. Keep a checkweigher downstream regardless: density drifts as product loses moisture out of the oven, and only a real weigh station closes that loop.


