Can a 3D camera see temperature or the effects of steam?

You have a hot or steaming product — trays leaving a filler, castings out of an oven, a wet washdown line — and one question on the table: will a 3D camera read the temperature zones, and will the steam turn into point cloud data? The short answers are no, and no. A structured-light 3D camera measures shape from a projected pattern; it is a rangefinder, not a thermometer, and steam never becomes geometry — it becomes noise. This explainer covers what these sensors actually measure, the two mechanisms by which steamy air corrupts a point cloud, and the arithmetic for pairing a thermal camera with your 3D sensor, so you can specify the combination with numbers behind it.

What a structured-light 3D camera actually measures

A structured-light camera projects a known pattern — coded stripes or fringes — onto the part and images it from an offset angle. Depth comes from triangulation: the apparent deformation of the pattern encodes the distance to every lit surface point. The output is X, Y and Z per pixel, usually with an intensity channel alongside.

Every value in that chain is a distance computed from where projected light landed. The intensity channel records how much visible or near-infrared light came back — a reflectivity measurement, not a heat measurement. Laser profilers swap the fringe pattern for a laser line and triangulate the same way; stereo cameras match texture between two views; ToF cameras time a pulse. None of those mechanisms touches temperature, a point worth weighing early when you choose the 3D technology.

A projected structured-light 3D camera of the class used on packaging and food lines:

Why temperature never reaches the point cloud

Silicon sensors respond from roughly 400 nm to 1000 nm: visible light plus a slice of near-infrared. A surface at industrial temperatures radiates far beyond that. Wien's displacement law puts the emission peak at 9.7 μm for a 300 K surface and 7.8 μm at 100 °C — long-wave infrared, where a silicon pixel does not respond at all. The 3D camera is blind to thermal radiation by construction, not by tuning.

A very hot part can brighten the intensity channel, but that glow is uncalibrated and shifts with finish, exposure and part colour; treat it as a coarse presence hint, never as temperature. When a specification calls for temperature zones on the product, the decision is closed: add a thermal camera working in the 8-14 μm band, and convert its counts to °C with a proper calibration — the step raw thermal counts to temperature walks through.

For a fixed camera over a line, the usual pick is a GigE thermal model. The LS-IRC-0044 is a 640 × 512 thermal camera covering 8-14 μm at 50 Hz on GigE.

Thermal images also carry far fewer pixels than a depth map — 640 × 512 against multi-megapixel point clouds — so the thermal layer, not the 3D camera, sets the smallest temperature zone you can read. That limit is worth computing before ordering; the worked example below does it.

What steam actually does to the point cloud

Steam never becomes point cloud geometry. The camera builds surfaces from points that hold still and reflect the pattern back; a vapour plume is semi-transparent, turbulent and full of droplets, so it fails on all three counts. What you get instead is noise, through two separate mechanisms.

Droplet scatter. Mist droplets reflect the projected pattern, and some of that return reaches the camera from ranges in front of the part. The cloud grows a flickering shell of false points that appears and vanishes frame to frame — spurious surfaces hanging in mid-air, obvious in a live viewer.

Heat-shimmer refraction. Air's refractive index falls as it warms, tracking density: heating air from 20 °C to 80 °C at fixed pressure lowers the index by about 5×10⁻⁵. Steep gradients over hot product and around the plume bend the outgoing pattern and the returning image — the same shimmer you see down a runway — leaving millimetre-scale ripple spread across real surfaces. The distortion grows with the length of air the light crosses, so long standoffs suffer most; read the 3D camera working distance and positioning guide before fixing one.

Fixes, in the order they pay off:

1. Move the viewpoint so the sight line crosses the plume as little as possible — shoot from the side once the part clears the hood rather than straight down through it.
2. Blow the plume out of the path: a cross airflow or purge curtain at the camera or the part is routine practice on filling lines.
3. Wait it out: on batch cycles a settle delay, or a photo-eye "steam clear" trigger before exposure, costs seconds and removes the mechanism.
4. Filter the remainder: drop points under the camera's confidence threshold — 3D camera confidence image values shows the flags — then re-measure critical dimensions on clean-air frames only, because ripple on valid surfaces survives filtering.

Worked example: thermal zones on a 200 mm part at 1000 mm

The case: a gantry over a conveyor with 1000 mm of standoff, chosen for clearance over the line. The product is 200 mm wide, and the smallest temperature zone worth reading is 20 mm. The thermal layer is the LS-IRC-0044; its horizontal field of view is 88.5° across 640 pixels.

The rule in words: the scene width equals twice the standoff times the tangent of half the horizontal field of view; one thermal pixel covers the scene width divided by the horizontal pixel count; a zone is readable once it spans at least three pixels.

The arithmetic:

• Half the field: 88.5° / 2 = 44.25°, and tan(44.25°) ≈ 0.974.
• Scene width: 2 × 1000 mm × 0.974 ≈ 1948 mm.
• One thermal pixel: 1948 mm / 640 ≈ 3.0 mm.
• The 200 mm part spans 200 / 3.0 ≈ 66 pixels.
• The 20 mm zone spans 20 / 3.0 ≈ 6.6 pixels, above the three-pixel floor.
• Smallest readable zone at this standoff: 3 px × 3.0 mm ≈ 9 mm.

Decision: at 1000 mm this pairing reads zones from about 9 mm upward, and the 20 mm zones pass with margin. Zones of 5 mm would fail; halving the standoff to 500 mm narrows the scene to 974 mm, gives 1.5 mm per pixel and a 4.6 mm floor — or fit a narrower-FOV lens. The 3D camera's lateral sampling at these distances is typically finer than 3 mm, so the thermal image, not the point cloud, sets the zone limit.

The camera resolution calculator runs the same arithmetic for any pixel count, field of view and standoff.

Pairing the thermal camera with the 3D sensor

Bolt both cameras to one rigid bracket and calibrate the extrinsic transform once — the workflow in multi-camera point cloud fusion with depth camera extrinsics applies directly, with a warmed calibration target in place of the second depth camera. Each thermal pixel then reprojects onto the point cloud along its ray, and zone temperatures land on geometry — the technique behind thermal RGB image fusion, where the temperature values must survive the colouring step.

For moving product, trigger both cameras from the same encoder pulse so the part sits in one pose in both images; free-running streams drift apart by whole zones at line speed. And not every job needs long-wave infrared: short-wave infrared cuts through light mist better than visible light, because longer wavelengths scatter less in small droplets. The MV-CI013-GS-NN is a fanless 1.3 MP SWIR camera on GigE. It measures reflected light, so it is a mist-penetrating 2D eye rather than a thermometer, and its image registers onto the cloud like any other modality — fusing depth, colour and infrared covers that step. Time-of-flight answers the same question here: the MV-DT01WMU is a ToF sensing camera, it times returning pulses, and droplets return those pulses too — so steam adds noise there just the same, and temperature stays invisible.

One mist-tolerant 2D sensor and one ToF unit for comparison:

Checks before you order

• Temperature as a number or as a gate: both need a thermal camera — simpler software for the gate, calibrated conversion for the number. No 3D camera covers either case.
• Sketch the plume against the sight line at its worst, the first tray after a fill. Choose the viewpoint that crosses the least plume, and budget a purge fan if crossing is unavoidable.
• Run the three-line arithmetic above at your own standoff and keep at least three thermal pixels per zone.
• Halving the air path roughly halves the refraction error; long standoffs over hot product are where ripple gets expensive.
• Steel glows visibly from about 525 °C: plan for lost pattern contrast on glowing parts, not for temperature data.
• One bracket, one calibration, one trigger source for both cameras; add a purge or heated window if washdown fogs the glass.

When the usual answer is wrong

• Thresholding the glow. The intensity channel brightens over very hot parts, and teams sometimes use it as a "hot part detected" signal. It works only for strong glow, is uncalibrated, and moves with exposure and finish; use it for coarse presence at most, and replace it with long-wave infrared for anything written into a spec.
• "Thermal 3D" in one housing. A product sold as a 3D thermal camera is a depth sensor and a thermal camera fused by the vendor. The fused output inherits both limits, so judge the depth accuracy and the thermal pixel count separately.
• Adding a visible camera. A standard area scan camera is equally blind — silicon stops near 1000 nm — and droplets scatter visible light hardest, so it solves neither problem.
• Fogged windows. Condensation on the camera window after washdown looks like permanent steam; the fix is a purge or heated window, because no exposure setting sees through the glass itself.

Take the standoff, the zone size and a photo of the plume, run the arithmetic, then start with 3D sensors in the camera finder, which filters both 3D and thermal models by interface and resolution, or open the system configurator. If the steam question is still open, contact us with your cycle time — a settle delay is often the cheapest fix on the list.