Review: Mini AMG8833 Thermal Camera

In our ceaseless quest to bring you the best from the cheaper end of the global electronics markets, there are sometimes gadgets that we keep an eye on for a while because when they appear they’re just a little bit too pricey to consider cheap.

Today’s subject is just such a device, it’s a minimalist infra-red camera using the 8 pixel by 8 pixel Panasonic AMG8833 thermal sensor. This part has been around for a while, but even though any camera using it has orders of magnitude less performance than more accomplished models it has remained a little too expensive for a casual purchase. Indeed, these mini cameras were somewhere above £50 ($70) when they first came to our attention, but have now dropped to the point at which they can be found for somewhere over £30 ($42). Thirty quid is cheap enough for a punt on a thermal camera, so off went the order to China and the expected grey parcel duly arrived.

The interface on this camera is about as simple as it gets.
The interface on this camera is about as simple as it gets.

It’s a little unit, 40 mm x 35 mm x 18 mm, constructed of two laser-cut pieces of black plastic held together by brass stand-offs that hold a PCB between them, and on the front is a cut-out for the sensor while on the rear is one for the 35mm OLED display.At the side on the PCB is a micro USB socket which serves only as a power supply. It’s fair to say that this is a tiny unit.

Applying power from a USB battery bank, the screen comes up with a square colour thermal picture and a colour to temperature calibration stripe to its left. The colours adapt to the range of temperatures visible to the sensor, and there is a crosshair in the centre of the picture for which the temperature in Celsius is displayed below the picture. It’s a very straightforward and intuitive interface that requires no instruction, which is handy because the device has none.

There’s something about a thermal camera, that as soon as you have one it is pointed at everything. People appear as red blobs, an extended hand sprouts barely discernible fingers, and a small pool of burning alcohol appears as a bright spot. Pointing it at a powered-up circuit board shows red blobs where each of the chips are, and easily allows low-resolution identification of the hot spots. Enough play though, just how good an instrument is it? A Prusa Mini heated bed makes a handy temperature reference, so it was heated up to 86 degrees for PETG printing and subjected to the camera. It’s a very rough and ready calibration, but the camera read it as 71.2 degrees and an infra-red thermometer at 78.4 degrees. Sadly I don’t have a thermocouple to hand to measure the Prusa directly, but I would be inclined to take any number generated by this camera to be very much on the low side.

The other use always shown for a thermal camera is to look at a building and see where it is losing heat. Taking the camera out in the night it could certainly see the windows of the house, but on a warm summer evening it probably had less to measure than on a freezing winter night with the heating on. It’s worth noting that the maximum distance it could discern a human at was consistently somewhere around 5 metres.

Having put the camera through its paces it was time to subject it to a teardown. Not an onerous task in this case, simply undo four screws and there’s the PCB. On it is little more than the sensor, the connectors, a microcontroller that defied our identification skills, and a few passives. It’s worth noting that the PCB is milled away between sensor and microcontroller, no doubt to stop thermal contamination. This is an astoundingly simple device.

The mini AMG8833 thermal camera then: It’s a functioning thermal camera at a very affordable price. So should you buy it? It’s fairly obvious that when compared to a more expensive thermal camera such as the FLIR Lepton-sporting tCam-Mini we reviewed recently, anything using an AMG8833 is a mere toy. So it’s senseless to berate it for not offering high resolution and other advanced features, because it cost a fraction of the price.

What it does give you is a usable way to evaluate the thermal signature of electronics and other devices, albeit with not entirely trustworthy calibration. It would have been nice to save images or had the USB connection offered some way to retrieve the sensor data, but even without that it delivers on expectations. Yes, it’s a toy, but it’s also a useful toy, and given that you’d probably spend more than the thirty quid it costs building your own version from modules I’d call it a diamond in the rough. I didn’t regret that particular AliExpress order, and I don’t think you will either.

29 thoughts on “Review: Mini AMG8833 Thermal Camera

      1. I imagine it would be usable for many things given the low cost of the unit, even at low res it would make for an interesting introduction to thermal imaging. As for the dubious calibration – how about measuring boiling water and ice then doing an average error compensation?
        Just a thought, I might be over-simplifying…

        1. I work with the calibration of thermal cameras, and can say that the calibration (for very accurate measurements, at least) is a lot more complicated than it looks at first, with distance, sensor temperature, material you are observing and even ambient humidity playing crucial roles in what the sensor outputs
          That said, i believe these are great for identifying hot/cold spots or getting a rough (very rough) reading of temperature on somethings, such as “is my hotplate heating at all?”
          Whereas if you need to see if a person has a fever or if a crucial component reaches a very specific temperature in a circuit, you’d be better off with a more expensive camera…

          1. “That said, i believe these are great for identifying hot/cold spots or getting a rough (very rough) reading of temperature on somethings, such as “is my hotplate heating at all?””

            And things like locating the preheated material for an automated seam welder. Rough actual temp, relative temp, and location within a well defined limit are what is needed (is it hot enough? where is it?) IIRC, the sensors are rotated 45deg to the travel to improve the positional resolution a bit.

          2. Thanks for pointing this out, Abo. It’s a real concern during the pandemic that people seem to think that a hacked-together thermal camera will reliably detect covid19 carriers. RTFM :-)

    1. I have one also, but packaged into a handheld unit which takes 3 AAA batteries and looks just like my Fluke thermal imager but smaller. It’s great and at the time was around £75. Used it to debug the heating system and far more sensative and safe finger for finding warm electrical/electrinic things in circuits.

      Some plumber friends have purchased them also as you don’t need to know the exact temperature of a radiator, but seeing the top half is cold clearly indicates it’s full of air.

      Yes, the resolution isn’t fantastic at all. But in the land of the finger/thermometer, the 8×8 imager is king. However, it looks like 32×32 imagers are crashing in price and hey’re much nicer. The fluke mentioned above is such, but was second hand. Fantastic for imaging where pipes are in walls and floors to avoid drilling into them. Hot is very easy, cold requires running it for a while. Really cool looking at under-floor heating. You can more or less make these out with the 8×8 too, just have to be twice as close.

      Do miss having access to a Flir that cost more than my Mortgage deposit though. That took great photos.

      1. Nope, just checked, the Fluke is 60×80 which is way more fun. Nevertheless, the little one with 8×8 is still king compared to burning your finger when poking a hot resistor.

        1. You can also look at the breaker and see if it is heating equally or only on one phase to spot loose terminal. I think it is one of those toys that are as bad as on a paper and as good as the owner creativity.

  1. There are two familiies of thermal imaging in common use – the first are the thermopile systems such as those in the AMG8833 – these are essentially little thermocouples in a grid. You can get pretty-okay resolution with these (I’ve worked with them for industrial imaging) and you can get some level of magnification with the ZnS lenses sold for use with laser cutters (be prepared for weirdness as the lenses are opaque to visible light). Frame speeds are limited by polling the thermopile array – I was getting about 10 – 15 FPS with one incarnation but it was irregular.

    The second are the microbolometers which are a little more complex, much more accurate and are at the heart of FLIR’s heavily defended patent fortress. These (plus really spendy ZnS lenses) can get you all kinds of resolution, frame speed etc. etc. but you’ll pay handsomely for it.

    You’ll also run squarely into Strategic Trade Authorization regulations – the inane business of assuming that someone can’t build a heat-seeking missile if the frame rate of higher resolution units is kept below 9Hz.

    So, with that said, certain units in Flir’s array of “consumer” thermal imaging products available on Amazon are directly seen as webcams and can be used as such – I’ve measured the frame rate at ~28 – 30 fps which is all you’ll need (unless you’re doing arcane lab stuff). They’re available on Ebay as there are now several generations.

    ZnS lenses can be constructed to fit, though there’s no bayonet/screw attachment point. They’re a bit more than the Flir One units, but are a lot easier if you’re doing signal processing on a laptop etc. Annoyingly you’ll also have to get a bicycle phone mount and adapt that to use a tripod with the thing, but you can’t have everything if you’re going cheap.

  2. https://www.youtube.com/watch?v=_lDm0RV9IV8

    That video explores using a little higher resolution and more expensive sensor but I think many of the conclusions apply here too, at least for electronics work.
    I would be curious to see if these could survive in the wheel well of a car to get telemetry on tire tread temps. If they could be tiled and the resulting set of images could be stitched together, that might add some additional possibilities too.

  3. I’m working on a handheld thermal camera using a 32 * 24 pixel sensor from Melexis. The project is at https://hackaday.io/project/181230

    I built it because I couldn’t afford a nice one from FLIR (and from these comments it sounds like they suck at supporting older devices!)

    I’ve also made it compatible with a range of substitute parts due to the component shortage… I actually designed this thing a year ago when I was adding insulation to my house, but the sensors weren’t available ANYWHERE until June this year.

    Bear with me, I’m still uploading all the info to hackaday.io :-)

  4. Is the sensor actually only 8×8 pixels? It looks like the images it’s generating are significantly higher resolution than that, maybe 32×32. Is that just interpolated?

Leave a Reply to akashuCancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.