Thermal Camera

Honest first, because it matters here more than usual: your phone has no thermal sensor. It cannot see heat. Nothing in this instrument measures temperature, and any figure it displayed in degrees would be an invention.

The reason is not a software limitation and no update will fix it. Thermal imaging works in the long-wave infrared, somewhere around eight to fourteen micrometres, and it needs a microbolometer — an array of tiny elements that physically warm up when that radiation lands on them. A phone camera is silicon. Silicon stops responding at about 1.1 micrometres, roughly a tenth of the way to where the interesting part of the spectrum begins, and there is an infrared cut filter bonded over the sensor removing most of what it could still have seen.

So what Instrument 12 does instead is take the ordinary visible-light image, measure how bright each part of it is, and map that on to the false-colour ramps real imagers use — the ironbow, the rainbow, white-hot, black-hot. A bright surface comes out looking hot. A dark one comes out looking cold. It is a convincing impression of an expensive instrument and it costs you nothing to know that it is an impression.

Why the numbers say REL

A false-colour picture makes a claim the moment it is orange in the middle, whatever the caption underneath says. So the withdrawal is on the picture rather than under it: the scale bar down the right-hand side of the frame is labelled REL — NOT TEMPERATURE, in the place a real imager prints its high and low in degrees, and it is there permanently.

Every figure the instrument produces is in relative luminance, nought to a hundred, where nought is the darkest thing the sensor registers and a hundred the brightest. That scale appears on the panel, on any plate you expose, and in your service record. There is no temperature field anywhere in this instrument, including in storage, so there is nothing for a later version of the site to quietly start filling in.

What is genuinely measured

Three things, and all three are real numbers taken off your own camera frame.

• The span. The window of brightness actually present in the frame, taken as percentiles of its histogram. This is the level-and-span control every thermal imager has, doing exactly the job it does on a real one: a picture stretched to the range that is present shows structure that a picture stretched to the full range cannot.
• The centre spot and the peak. A real imager prints the value under its reticle and boxes the hottest pixel in frame. This one prints the value under the reticle and boxes the brightest region — same instrument, honest unit.
• The differential. The one that justifies the whole instrument. See below.

The reference frame, which is the reason to keep it

Press hold reference and the instrument stops showing you brightness and starts showing you change. Everything that has not moved since the moment you pressed it goes to black. Anything that has lights up, and the panel tells you how far it moved and what fraction of the frame it covers.

This is not a trick, and it is not an approximation of thermography — it isthermography’s most useful operation, done in the band this device can actually see. Differential work is what a real thermographer spends their time on: photograph a wall, come back, look at what changed. A single thermal frame tells you a surface is nineteen degrees; the difference between two frames tells you something happened.

Put the phone down somewhere it can see the room, hold a reference, and step back. Everything still is black and you are the only thing in the frame. It is the most genuinely unsettling thing on this instrument and every pixel of it is a real measurement.

A palette is a lookup table and nothing more. It contains no information: the same frame through two palettes is the same measurement twice. Which one to use is a question about your own eyes.

• Ironbow. The one everybody pictures. Black through violet and red to white. Best general palette because human vision discriminates hue better than lightness, so a gradient across several hues shows more structure than a grey one does.
• Rainbow. More hues again, and consequently more contrast in the middle of the range and more false edges. Useful when you are hunting a boundary; misleading when you are looking at a smooth surface, because it invents bands that are not there.
• White hot and black hot. Greyscale, and the two ways round. Real operators use these more than they use the colourful ones, because a monochrome image is the one where you can trust that a change in the picture is a change in the reading.
• Image intensifier. Not a thermal palette at all. This one does what a night sight does — throw away colour, put the gain up, and accept the grain that comes back with it. The green is the phosphor of an intensifier tube. The grain is your camera’s own sensor noise, amplified rather than added: nothing on this instrument invents texture.

And the narrow span

The span control has a third setting that collapses the mapped window to a few levels either side of the frame’s median. This is a real operation with the gain wound past all sense, and what it amplifies is the sensor’s noise floor — which in a dark room is most of what the sensor is producing.

The result writhes, and it suggests shapes, and it will keep suggesting them for as long as you watch. That is worth knowing about rather than being protected from, because it is exactly the mechanism behind a large fraction of what gets posted as evidence from cheap thermal and night-vision equipment. The shapes are in the sensor. They are in everybody’s sensor. There is a whole note on cameras that goes further into this.

Throw the switch and let the correction run. The shutter closes across the sensor, holds, and opens. On a real uncooled imager this is a flat-field correction — it photographs the back of its own shutter and subtracts the result — and it is the click a thermal camera makes every few minutes of its life. Here it takes about a second and a half, and then the instrument walks its palettes so you know they are there.

Leave it on automatic span to begin with. Wide is honest about how little range an ordinary room has and is mostly useful for showing you that. Narrow is the noise.

Then hold a reference and do not move. This is the whole instrument. Set the phone against something solid, press hold, and step out of shot. The picture goes black because nothing is changing. Everything that happens afterwards is a real measurement of a real change.

Expose a plate when something is worth keeping. The picture, a bezel and a typed caption on paper, with the palette, the span, both figures and the honest line. Two presses: expose, then share. The second one hands over a file that already exists, because the share sheet needs a gesture and encoding an image is not instantaneous.

What will move the picture, in descending order of how often

• Your own hand, if you are holding the phone. It is warm, it is close, and in visible light it is also bright. It will be the peak in almost every frame.
• A window, a lamp, a screen, or anything else emitting light. These are not hot, they are lit, and on this instrument that is the same thing.
• A pale wall against a dark one. Emissivity does this to real imagers too, for different reasons and with the same result: a surface property masquerading as a temperature.
• The camera’s own automatic exposure. It is adjusting continuously, and on a differential reading a global exposure shift reads as the entire room changing at once. If the whole frame lights up, that was the camera, not the room.

That last one is worth dwelling on, because it is the honest limitation of this instrument and there is no version of it that does not have that limitation. A real imager holds its gain fixed and a phone camera does not, so a differential reading here is a measurement of the change in the picture rather than of the change in the room. Most of the time those are the same thing. When the whole frame moves together, they are not.

Ghost Detector Lab is an entertainment site. See the disclaimer, the privacy notice, and the method for every instrument on the bench.