A laser triangulation sensor projects a spot or a line onto a surface and watches it from an angle. Move the surface towards the sensor and the image of the spot shifts sideways on the detector. Calibrate that shift and you have a distance measurement, with no contact, no moving parts and a measurement rate limited mostly by the camera.

It is the workhorse of industrial dimensional measurement. It is also the method where the obvious way to buy accuracy, a better detector and finer interpolation, stops working earlier than most engineers expect. This article is about where the limit actually sits.

The geometry in one page

Laser, surface and detector form a triangle. The laser axis and the observation axis meet at the triangulation angle Θ. A displacement of the surface along the laser axis moves the spot image across the detector by an amount that follows from that angle and the magnification of the imaging optics:

A larger angle gives more lateral shift per millimetre of depth, so better depth resolution. It also means the observation path is more oblique, so steep flanks and holes go into shadow sooner and the sensor loses the signal. Every triangulation design is a choice on that line, and the datasheet specification you compare between suppliers is a consequence of where they chose to sit.

A second geometric detail that separates working sensors from breadboard setups: because the object plane is tilted relative to the optical axis, the image plane has to be tilted too in order to keep the spot in focus across the whole measuring range. That is the Scheimpflug condition, and in a commercial sensor the detector is physically mounted at an angle because of it.

Speckle sets the floor

Here is the part that surprises people who come to triangulation from electronics. The accuracy limit of a triangulation sensor on a technical surface is not the pixel pitch and not the noise of the detector. It is speckle.

Any surface that is rough compared to the wavelength scatters coherent light into a random interference pattern. The spot image on the detector is not a clean intensity profile but a granular one, and the centroid of that granular profile differs from the true centre by a random amount. The sensor reports that error as distance. The same speckle that carries the signal in speckle contrast imaging is pure noise here.

Dorsch, Häusler and Herrmann derived the resulting limit in 1994, and it has held up since:

Wavelength	Θ = 30°, sin u = 0.05	Θ = 30°, sin u = 0.10	Θ = 45°, sin u = 0.05
405 nm	2.6 µm	1.3 µm	1.8 µm
520 nm	3.3 µm	1.7 µm	2.3 µm
660 nm	4.2 µm	2.1 µm	3.0 µm
785 nm	5.0 µm	2.5 µm	3.5 µm

All figures assume full speckle contrast, C = 1. Read the three columns as the three levers you have: wavelength, aperture and triangulation angle. Note that none of them is the detector.

This does not make speckle disappear. Averaging several exposures, moving the part, or wobbling the beam all reduce C further, at the cost of measurement rate. But it does mean that money spent on a narrow-line source buys nothing in this application, and may cost you accuracy.

Point or line: two problems

The term covers two families of instrument that share the geometry and differ in almost everything else.

	Point triangulation	Line triangulation
What it measures	distance at one location	a height profile across the line
Detector	line sensor or PSD	area sensor
Optics after the laser	focusing lens	line generator, commonly a Powell lens
Typical use	displacement, thickness, edge and gap measurement	weld seam tracking, profile and 3D scanning
What limits the rate	sampling and interpolation	sensor readout and profile extraction

The optical requirement differs where it matters for the source. Point triangulation needs a small, well-defined focus, so beam quality translates directly into the smallest feature the sensor can resolve laterally. At 50 mm focal length and a 2 mm input beam, a diffraction-limited Gaussian focus comes out around 13 µm at 405 nm and 17 µm at 520 nm, measured as the 1/e² diameter, and scaling with wavelength as expected.

Line triangulation additionally needs the intensity to be even along the line. A cylindrical lens spreads a Gaussian beam into a line that is bright in the middle and weak at the ends, which costs you the outer part of your field. A Powell lens redistributes the energy and holds the profile closer to uniform. That is a question of the optics in front of the source rather than of the source itself, but it decides how much of your field of view is usable.

Why blue won the hard cases

Red diodes at 660 nm were the default for years, because they were cheap and silicon sensors are sensitive there. Blue and violet have taken over the difficult applications for two distinct reasons, and it is worth keeping them apart.

Glowing parts. A hot surface emits its own light, and at the temperatures of rolling mills and forges that emission swamps a red laser. The Planck spectrum falls steeply towards the blue, so moving the laser away from the emission peak buys an enormous amount of contrast. At 1000 °C the surface radiates roughly 4,000 times more per unit wavelength at 660 nm than at 405 nm. At 1200 °C the advantage narrows but is still around a factor of 1,000. The ratio is steep in temperature, so treat these as orders of magnitude rather than as figures to design against. A narrow interference filter in front of the detector then removes most of what remains.

Translucent materials. On skin, many plastics, ceramics, marble and some composites, light does not reflect at the surface. It enters, scatters inside and leaves again from a region around the entry point. The spot the sensor sees is therefore broader and its apparent centre sits deeper than the real surface, which shows up as a systematic distance error rather than as noise. In most of these materials absorption is weaker towards the red and near infrared, so penetration depth grows with wavelength. A violet or blue spot stays closer to the surface and the error shrinks.

Wavelength	Where it is the right choice	Where it is not
405 nm	glowing metal, translucent plastics, skin, the finest resolution	sensor sensitivity is lower, and some paints fluoresce under violet
450 to 520 nm	general purpose on difficult materials, good balance of resolution and detector response	little against it; a common choice on newer designs
660 nm	cooperative surfaces, cost-sensitive designs, good detector sensitivity	glowing parts, translucent materials
785 nm and above	where the beam must stay invisible, or ambient light in the visible is strong	worst resolution of the set, deepest penetration into translucent parts

Fluorescence deserves a flag. Some coatings, adhesives and plastics fluoresce under violet illumination, which produces a broad glow around the spot and degrades the centroid. Where that happens, a longer wavelength or a filter matched to the laser line solves it, and it is worth testing on the actual material rather than assuming.

What the surface does

Three surface effects account for most of the difference between a datasheet figure and what a sensor achieves in a machine.

• Specular surfaces. A polished part reflects the beam away from the observation aperture and the sensor sees nothing, until the angle happens to be right and it sees far too much. Tilting the sensor, using two observation channels, or accepting a matte reference feature are the usual answers.
• Steep flanks and edges. Near an edge, part of the spot sits on one surface and part on the other. The centroid lands between them and the sensor reports a distance that exists nowhere on the part. This is a geometric artefact, not noise, and it does not average away.
• Varying reflectance. A black-to-white transition inside the spot biases the centroid towards the bright side. Sensors with exposure control handle the dynamic range, but not the bias within a single spot.

Ambient light matters less than people expect, provided the design takes it seriously: a narrow bandpass filter matched to the laser line, short exposures and sufficient laser power together suppress most of it. What they do not suppress is light at the laser wavelength itself, which is why another triangulation sensor pointed at the same area is a more serious problem than a window.

What to specify

Parameter	What decides it	Guidance
Wavelength	material, surface temperature, detector sensitivity	405 to 520 nm for glowing or translucent parts, 660 nm where the surface is cooperative
Coherence	speckle contrast, which enters the accuracy linearly	the one case where a broader line helps; do not pay for narrow
Beam quality and focus	lateral resolution and depth of field	TEM00 for a small spot, but a large depth of field needs a larger waist, so it is a trade
Power stability	centroid bias and exposure control	RMS over the measurement interval, stated with its bandwidth
Pointing stability	drift of the spot translates directly into apparent distance	specify over the operating temperature range; this is often the dominant drift term
Modulation	ambient light rejection, multi-sensor operation, fast gating	direct modulation of a diode is simpler than external modulation of a DPSS source

Pointing stability is the specification most often overlooked and the one that most often explains a slow drift in a production measurement. The sensor cannot distinguish between the beam moving and the part moving. Where the figure matters, it needs to come from a datasheet with its measurement conditions attached, not from a general statement about the product family.

Frequently asked questions

What accuracy can a laser triangulation sensor reach?

On a rough technical surface at full speckle contrast, the limit lands in the low single-digit micrometres for typical geometries, and that is before mechanical and thermal effects. Full contrast is the worst case, so read it as a ceiling on the uncertainty rather than as the accuracy you will see. Sensors quoting far better figures are often measuring a cooperative surface, averaging many readings, or quoting repeatability rather than accuracy.

Why is a blue laser better for hot parts?

Because the part’s own thermal emission is far weaker in the blue. At 1000 °C the surface radiates roughly three to four orders of magnitude more at 660 nm than at 405 nm, so a blue laser plus a narrow filter still gives a usable spot where a red one disappears.

Does a narrower linewidth improve a triangulation sensor?

No, and it can make things worse. Accuracy is limited by speckle contrast, and a broader spectrum lowers that contrast. This is the opposite of interferometry and of vibrometry.

What is the difference between point and line triangulation?

Point triangulation measures distance at one location with a focused spot and a line sensor. Line triangulation projects a line, usually through a Powell lens, and reads a full height profile from an area sensor.

Why does my sensor read wrong near an edge?

Part of the spot sits on each side of the edge and the centroid lands between them. It is a geometric artefact of the finite spot size, so a smaller spot narrows the affected zone but does not remove it.

Can two triangulation sensors look at the same area?

Not without care. A bandpass filter does not separate two lasers at the same wavelength. The usual answers are different wavelengths, time-multiplexed operation, or modulating the sources and detecting synchronously.