Temperature vs position line profile from radiometric pixel data

Your tube, seal or heater strip carries a temperature spec along its length, and one spot reading cannot tell you where it drifts. A radiometric thermal camera already holds one calibrated temperature in every pixel, so the profile you want is one drawn line away: draw the line, export it, scale the pixel index into millimetres, and temperature versus position falls out as a plain array of numbers. This page gives the scaling arithmetic, a worked profile across a 100 mm tube at 0.42 mm per pixel, what the exported file contains, and how a script or controller charts and checks it automatically without an image ever being decoded.

What is a temperature vs position line profile?

A line profile is the row of calibrated temperature values a radiometric camera stores for the pixels under a drawn line, plotted against distance along that line. A 240-pixel line across a 100 mm tube is one temperature value every 0.42 mm.

Because the camera is radiometric, every pixel already carries an absolute temperature. There is no color scale to decode and no precision lost reading a palette back into numbers — that trap belongs to color-mapped images. The profile keeps the calibrated values in pixel order and adds exactly one thing: position.

The three readouts a thermal camera supports answer three different questions:

How do you turn pixel index into position in millimetres?

Multiply the pixel index by the scale of the scene. Millimetres per pixel is the object width the line spans divided by the pixels it covers, so 100 mm spanned by 240 pixels works out to 100 / 240 = 0.4167, rounded to 0.42 mm per pixel.

Measure the span at the part's own plane, with a ruler laid beside it or a known length machined into the fixture. Detector-level pixel pitch — the physical centre-to-centre distance between pixels on the sensor — does not change with distance. The step you chart does: double the working distance and every pixel step doubles. That is why the working distance and spot size guide treats the two as one decision. Keep the line on a single plane; pixels crossing parts nearer or farther than the calibration plane inherit the wrong scale.

Worked example: a heater tube is 100 mm long and the drawn line crosses exactly 240 pixels end to end, so one pixel is 100 / 240 = 0.4167 mm. The spec allows ±20 °C around 950 °C between 10 mm and 90 mm, which is pixel indices 10 / 0.4167 = 24 through 90 / 0.4167 = 216. The export shows 40 consecutive pixels starting at index 120 reading below 930 °C. First cold pixel: 120 × 0.4167 = 50.0 mm. Band length: 40 × 0.4167 = 16.7 mm. The fault therefore sits 50-67 mm from the line start — a location in millimetres, not somewhere on the tube.

What does the exported profile file contain?

A line export is plain tabular data in pixel order — one absolute temperature per pixel. A 240-pixel line arrives as 240 numbers that load into a one-dimensional float array of about 2 KB as text, with no image decoding at all.

Export tools write the values as one row or one column of comma-separated text. A spreadsheet lays the row across the screen, so an unrotated line reads as a trace lying on its side rather than a chart; rotate it, or load it into code in the first place. Three export variants exist in practice:

• Temperature CSV: values already carry the emissivity, distance and window corrections. Chart it directly.
• 16-bit radiometric image: one number per pixel with the calibration stored beside it. The radiometric JPEG guide shows what to preserve when saving, and the high-bit-depth export guide how to keep every level.
• Raw counts: uncorrected sensor output. Convert with a calibration before the array means temperature — the raw counts to temperature guide walks the conversion.

This is not the same job as rendering the raw pixel array as a picture: that guide displays the 2-D array as an image, while a profile reduces one row of it to temperature versus position. A full 640 × 512 radiometric frame is 640 × 512 × 2 bytes ≈ 0.7 MB. Profiles are light to log every cycle and full frames are not.

How accurate is the temperature along the line?

The values are only as good as the emissivity and window settings behind them: factory calibration for an uncooled LWIR camera is typically ±2 °C or ±2 % of reading. A wrong emissivity on a shiny surface can move a reading by tens of degrees.

Set emissivity per surface, not per material family. An oxidized steel wall sits near 0.80-0.90 in the 8-14 μm band, while polished metal at 0.1-0.2 reflects its surroundings and misreads badly. Sighting into an open tube end helps, because the cavity drives the effective emissivity toward 1. Where the line crosses a sight window, enter the window's transmission in the compensation settings — window materials absorb part of the 8-14 μm band and re-emit some of their own, and the infrared window material guide compares the options.

A furnace wall hotter than the part reflects onto the reading; view at an angle or shield the sight path. Pixel noise is not the limit: uncooled LWIR industrial cameras are commonly specified near 0.03-0.05 °C NETD, so gradients of a few degrees sit far above the floor — the NETD and lens guide covers that trade.

How fast can the profile refresh?

The camera link never sets the pace: a 50 Hz LWIR camera delivers a fresh frame every 20 ms, so the profile rate is set by the trigger and export step. In a slide-in fixture that means once per mechanical cycle.

In a furnace fixture the camera head physically slides to the window, shoots and retracts, so cycle times of several seconds come from motion, not data. One profile is 240 values, or 240 × 8 = 1,920 bytes as 64-bit floats — trivial to process. Poll the export folder every 250-500 ms, grab the newest file, and the chart updates within about a second of the shot.

Can a controller or script chart it without the vendor software in the loop?

Yes — the export is an ordinary file on a disk, so anything that can watch a folder, read text and plot an array can own the profile end to end. The chain is: trigger the save, load the newest CSV, scale it, check it, chart it.

1. Trigger. The controller closes a digital output, or the acquisition code saves the frame, once the slide reaches the window.
2. Watch. Poll the export folder every 250-500 ms and take the newest file by creation time.
3. Load. Read the line into an array of n floating-point values, one per pixel.
4. Scale. Multiply each index by the mm-per-pixel factor from the worked example.
5. Check. Count pixels outside the 930-970 °C band across the in-spec span and compare with the reject limit.
6. Report. Plot temperature on the y axis, position in mm on the x axis, draw the two limit lines, and hand the pass flag and the worst position back to the controller as values it can use.

Keep the corrections — emissivity, distance, window transmission — in whatever saves the frame. The reading and plotting chain consumes finished temperatures and never changes them, and moving processed files to a done folder guarantees a shot is never charted twice.

How many pixels do you need across the part?

Work backwards from the smallest feature you must locate: allow 3-5 pixels across it, so a 1 mm feature sets the scale at 0.2-0.33 mm per pixel. A 120 mm field of view then needs a line of 360-600 pixels.

Size the field before the camera: a 100 mm tube with 20 % margin needs a 120 mm field of view. A 640-pixel camera across it samples every 120 / 640 = 0.19 mm per pixel — fine enough for the 1 mm rule. Where the field of view is fixed at the lens, working distance does the tuning instead. The LS-IRC-0044 is a 640 × 512 LWIR thermal camera with a printed 88.5° × 73.2° field of view, a GigE interface and a 50 Hz frame rate.

That 88.5° horizontal angle spans 2 × 200 mm × tan(44.25°) ≈ 390 mm at a 200 mm working distance, so a full-width line samples every 390 / 640 ≈ 0.61 mm; halve the distance and the step halves with it. For a tighter field, fit a longer thermal lens — the LS-SWL-0006 is a 17 mm F1.2 LWIR thermal lens covering a 1/1.2" format with an 11-16 mm image circle.

For a compact cell, the LS-IRC-0016 is a 640 × 512 LWIR camera module with USB3 output; the profile chain stays identical, only the connection differs.

Check your own numbers with the calculator below — it turns a field of view and a required detail size into the pixel count to filter for — then narrow the camera finder to the infrared cameras, and pick matching glass from the thermal, SWIR and UV lenses.

When is a line profile the wrong tool?

Change tools when the question becomes two-dimensional or the answer is a single number. A full-frame export finds an unknown hot spot, and a spot pyrometer is the simpler instrument when one fixed point is the entire specification.

Emissivity that varies along the line — oxide patches, joints, two materials in one row — mixes surface changes with temperature changes, and the profile then needs material boundaries before it means anything. If the defect could sit anywhere on the surface, export the full frame first and profile the region you find. If the part is one point of interest on a fast control loop, a pyrometer answers in milliseconds — the thermal camera versus spot pyrometer guide lays out that decision.

What should you check before wiring the profile into production?

Six checks separate a profile that reports the truth from one that quietly scales wrong. Measure the span, count the pixels, size against the feature, record the corrections, define limits in millimetres, and trial-run the folder chain before the alarm goes live.

• Span measured at the part's plane. Every position scales from it; a 5 % span error shifts every reported location by 5 mm on a 100 mm tube.
• Pixel count across the line. Sets the sampling step: n pixels means n temperature points, never more.
• Scale against feature size. Keep smallest feature ÷ 4 below the mm-per-pixel step to put 4 pixels on any edge you must catch.
• Corrections recorded once. Emissivity, distance and window transmission written down means a changed reading points at the process, not the settings.
• Limits defined in mm, not pixel indices. A camera swap or a distance change then leaves the pass band untouched.
• Twenty trial profiles watched end to end. Bounds the noise and the folder-watcher behaviour before anything automatic acts on the numbers.

Send the tube length, working distance and the smallest feature you must catch, and we will check the sampling budget with you before you order. Start from the infrared cameras shortlist or the lens category above, and contact reaches the applications team directly.