As an Amazon Associate, Akermin earns from qualifying purchases. Product links on this page are affiliate links.
Accuracy in a thermal camera is a chain, not a single number. The quoted figure, typically plus or minus 2 percent of reading or 2 degrees Celsius, applies only when emissivity, distance and ambient conditions are set correctly. Resolution and thermal sensitivity decide whether you can see the target at all; accuracy decides whether the number attached to it is meaningful. For most building, electrical and HVAC work, a 320 by 240 sensor with a stated NETD and a manual emissivity control is the practical threshold. A USB-C phone camera such as the AMPBANK M192 can be genuinely useful for finding anomalies, but its absolute readings depend on the same corrections as any other camera, and no handheld camera replaces a contact thermometer when a number has to stand up in a report.
The accuracy figure on the box covers less than you think
Every thermal camera listing quotes an accuracy, usually as a percentage of reading plus a fixed offset. The MiLESEEY TR20 states plus or minus 2 percent; the AMPBANK M192 states a maximum of plus or minus 2 degrees Celsius or 2 percent. Read that as the error of the detector and its internal reference, measured under laboratory conditions with the camera pointed at a blackbody of known temperature. It is not the error of your reading on a live electrical panel.
Three other errors sit on top of it. Emissivity is the first: a surface that reflects rather than emits infrared will read low, sometimes by tens of degrees. Distance is the second: the spot the camera averages grows with distance, so a small hot component fills less of the detector and its heat is diluted by the cooler surroundings. Ambient reflected temperature is the third: a cold window or a warm radiator behind you changes what the camera sees. None of these appear in the accuracy specification, and all of them can exceed it.
This is why a camera with a good accuracy figure and no emissivity control is less useful than one with a slightly worse figure and a proper settings menu. The AMPBANK listing mentions adjustable emissivity, ambient temperature and distance settings inside the app, which is the right set of controls for the price class. A camera that hides those settings is quoting an accuracy it cannot deliver in the field.
Resolution and NETD decide whether you get a reading at all
Accuracy is meaningless if the target is too small to fill a pixel. A 320 by 240 sensor, as in the MiLESEEY TR20, gives roughly 76,800 detector elements; a 192 by 192 sensor gives about 36,800. At a fixed distance the lower-resolution camera spreads each element over a larger patch of wall or circuit board, so a small hot joint is averaged with its cooler neighbours. The reading falls, and the fault can disappear.
NETD, noise equivalent temperature difference, is the smallest temperature change the sensor can distinguish from its own noise. The TR20 states under 40 millikelvin; the DCTOPT states under 50 millikelvin. Lower is better, and the difference matters most when you are looking for a few degrees of warmth in a wall or a duct. A camera with poor NETD will show a noisy image where the real pattern is buried.
Super-resolution modes, such as the 192 to 384 enhancement on the AMPBANK or the 240 by 240 mode on the DCTOPT, interpolate between detector elements. They make the image easier to read and can help locate a feature, but they do not create new temperature information. Treat them as presentation, not as an accuracy improvement. For more on how pixel count translates into a usable measurement, see thermal camera resolution explained.
What the quoted specifications actually compare
The table below puts the stated figures side by side. Where a listing is silent, the cell says so. No figure has been estimated.
| Feature | MiLESEEY TR20 | Bosch GTC600C | AMPBANK M192 | DCTOPT |
|---|---|---|---|---|
| Detector resolution | 320 x 240 | Not stated in the listing bullets | 192 x 192, enhanced to 384 x 384 | 240 x 180, enhanced to 240 x 240 |
| Frame rate | 25 Hz | Not stated in the listing bullets | 25 Hz | 25 Hz |
| Thermal sensitivity | Under 40 mK | Not stated in the listing bullets | 40 mK | Under 50 mK |
| Temperature range | -4 to 1022 F | Not stated in the listing bullets | -4 to 842 F | Not stated in the listing bullets |
| Accuracy | Plus or minus 2 percent | Not stated in the listing bullets | Plus or minus 2 C or 2 percent | Not stated in the listing bullets |
| Display | 2.8 in LCD, 480 x 640 | Not stated in the listing bullets | Phone screen via app | 2.8 in LCD |
| Storage | Not stated in the listing bullets | 500 images, 350 voice notes | App dependent, not stated | 4 GB card, over 10,000 images |
Note: the Bosch listing publishes image storage and voice note counts but no detector or accuracy figures, so it cannot be compared on measurement grounds from the supplied text alone.
Where a handheld camera stops being the right instrument
A thermal camera measures surface temperature by receiving infrared radiation. It cannot see inside a wall, inside a conduit, or through a metal enclosure. It cannot measure air temperature directly, and it cannot measure the temperature of a shiny metal surface without a correction that most users will not apply. If your question is whether a pipe is hot, a contact thermometer or a clamp probe gives a better number. The comparison is set out in thermal camera versus infrared thermometer.
If your question is what is behind a surface rather than how warm the surface is, a camera shows the thermal symptom but not the cause. A borescope or a moisture meter answers a different question; see thermal camera versus borescope. And if the surface is bright metal, the reading may be dominated by reflection rather than emission, which is covered in why shiny metal can mislead a thermal camera.
The honest answer to many accurate thermal camera questions is that the camera finds the anomaly, and a second instrument confirms it. That is not a weakness. It is how the tool is meant to be used.
Choosing between the instruments on this page
If you need a self-contained handheld with a stated accuracy and a wide range, the MiLESEEY TR20 publishes the most complete measurement specification of the four. If you want to record findings with voice notes on a job, the Bosch GTC600C is built around that workflow, though its listing does not publish detector or accuracy figures. If you already carry a USB-C phone and want the lowest-friction way to look at a wall or a panel, the AMPBANK M192 is the practical entry point, with the caveat that its readings live inside an app and depend on the settings you enter. The DCTOPT sits between them, with a stated sensitivity and on-board storage for image analysis.
None of these cameras is a calibration instrument. If a temperature has to be defensible, check the camera against a known reference and record the emissivity setting you used. The site’s review of a comparable handheld, the Seek Thermal thermal imaging camera review, shows how much of the practical experience comes down to the settings rather than the sensor.
The reading you can defend
An accurate thermal camera is one whose limits you understand and whose settings you control. The detector accuracy is the smallest part of the error chain. Emissivity, distance and reflected temperature decide whether the number on screen means anything. Buy for the resolution and sensitivity you need to see the target, then learn the corrections that turn a picture into a measurement. Where a listing does not publish a figure, treat that as a gap, not as a promise.
Frequently Asked Questions
Is plus or minus 2 percent accurate enough for electrical inspection?
It is accurate enough to find a hot joint and rank it against its neighbours. It is not accurate enough to certify a connection temperature to a standard without correcting for emissivity and distance. For fault finding, the pattern matters more than the absolute number.
Do super-resolution modes make a camera more accurate?
No. They interpolate between detector elements to make the image clearer. The underlying temperature information is unchanged. A 192 by 192 sensor enhanced to 384 by 384 still has 192 by 192 of real measurement data.
Why does my camera read low on shiny metal?
Shiny metal has low emissivity, so it radiates little of its own infrared and reflects the surroundings instead. The camera receives a mix and reports a temperature that is often closer to the reflected background than to the metal. Matte tape or a correction factor is needed.
Can a phone thermal camera be accurate?
The sensor can be, within its stated figure, but the reading depends on the app settings. If emissivity, distance and ambient temperature are not entered, the number is an estimate. The AMPBANK listing states these controls exist, which is the minimum for serious use.
What does NETD tell me that accuracy does not?
Accuracy describes the error on a known target. NETD describes the smallest temperature difference the sensor can resolve above its own noise. A camera with good accuracy but poor NETD will struggle to show a faint thermal pattern at all.
When should I use a contact thermometer instead?
When the surface is shiny, when the target is too small to fill a pixel, or when the number has to be defensible in a report. A contact probe removes emissivity and distance from the error chain entirely.