One place where the camera could do with some improvement is the visual camera supporting the thermal sensor which has a 640 x 480 pixel resolution. Most infrared thermal imaging cameras will have at least a 2 MP visual camera. Third, the start time of over 20 seconds. And when changing between high and low gain (>150°C) you can wait another 20 seconds... It is not easy to find out except to ask UNI-T what the physical resolution of the microbolometer is. Manufacturers sometime obscure that information deliberately. Whenever low-cost manufacturers introduce a high-end thermal camera in their portfolio, I’m always intrigued. This is the moment where they bring the fight to the big manufacturers and try to bring more to the table in terms of numbers. There’s an unwritten rule: The flagship of one company will be a better offering when compared to the mid-range of another. Why? Because the other company has their own flagship as well which costs even more and thus the mid-range can’t go that far in terms of specs because it will make the flagship redundant. It’s marketing 101.

The boot time is around 20s (mostly there is a progress bar) and the switching between the low and high gain takes also around 20-25s (there is no progress bar here). High gain is -15-150C, low gain 150-550C. Now let’s look at the camera’s temperature measurement performance. First of all, it has to be the sensitivity since that is instrumental to defining image quality. This camera has a 50 mK NETD which means it’s not bad at all. It means it can spot a temperature difference of down to 50 mK. The lower the number here the more detailed the image will look because the camera is sensitive to even the slightest differences in temperature. I don't think you could set the temp range manually either? That would be a no go for me too. But everyone has different needs! All the thermal images that you capture will be stored on the 16GB microSD card which is provided with the camera. That is more than enough and you basically won’t have to delete any photos throughout the camera’s lifetime. Personally, the device would drive me insane. But for the price it's hard to beat. You get what you pay for. This becomes quite clear again here. The biggest problem with handheld thermal cameras is the built-in processors, and so is the cost. This then leads to such effects as a delay in the display when panning (despite 25Hz). Or how long the device boots. Mobile phone models are better because the smartphone can do all the work, including an additional strong image improvement. Or you have to spend more money.If a camera is received and it is displaying dead pixels on its display..... it has suffered pixel failures since original calibration and should be sent for a new NUC and dead pixel map creation. Dead pixels are not truly a “Fault” in terms of a camera containing them..... they are a fact of life that the image processing system is designed to cope with. I suspect that the refresh rate is not reaching the spec (or because it's < it's not even close to it) and there is also some lag if I move the camera but for my use it's OK. The thermal imager camera is also equipped with a 5000 mAh Li-Ion battery which will ensure up to 6 hours of continuous operation.

It is not that easy to discover the dead pixel count without entering the cameras engineering modes or accessing the dead pixel map. A dead pixel map often exists as an image file containing all the pixels present on the FPA but highlighting those that are market based. The image processing stages read the dead pixel locations out of the image file. Gaining access to the dead pixel map is not a simple task on many cameras unless access can be gained to the operating system and configuration files.It is difficult for a user to establish how many non functioning pixels are present on a thermal sensor FPA for the reason that you detail. All FPA’s will contain pixels that are either faulty or produce an output that falls outside of the acceptable specification. The image data from the ROIC is normally RAW and subsequent image processing stages create the Non Uniformity Correction and Flat Field Correction tables that both capture ‘out of specification’ pixels and try to correct other pixel outputs to achieve a good Flat Field output. The dead pixel map is created by the NUC ‘calibration’ process carried out at the factory. Any pixel that produces an unacceptable pixel output value is marked as ‘Dead’. I can't say much about accuracy as I don't have anything similar to compare but it agrees with my other measurement methods (temperature probe of DMM included as well). Whilst dead pixel concealment is very effective, in applications where EVERY pixel output is being analysed, such as in some science applications, it is important for the user to know which pixels are not truly active and their data should be discounted from the results. This is limited to science applications though and not really an issue with general camera use.

Then there is the temperature range and things look fantastic here since the sensor can measure any temperature between -5 F to 1022 F (-15 C to 550 C), a range which is about 30% wider than what you would expect from other thermal cameras in this price range. Its thermal accuracy is on par with what’s expected out of thermal sensors these days, +/- 2 C.Once the Dead Pixel map has been produced the image processing stages of the camera do their best to disguise those pixels from the users view. This is relatively easy in most cases as the values of surrounding pixels may be used to create an average value to replace that of the dead pixel position in the array. Life becomes a little more challenging when a cluster of dead pixels or a dead column is detected. A cluster can cause a dead spot in the image displayed that cannot be concealed by the image processing and a dead column can be a challenge to hide from the human brain that sees pixel correlation and recreates the defective line in some cases. For these reasons a thermal FPA sensor specification normally states that the FPA shall not contain more than a certain number of pixels in a cluster and adjacent to each other. Dead Columns may also be a reason to reject an FPA. It probably won't see much scientific usage so should there be any pixel errors it's not my concern (the display has none, and sensor pixel fault is hard to detect as probably similar interpolating algorithms are used as with photography cameras so I don't know if it has one). It drains battery quite fast (or I've played a lot with it without noticing) but it seems to be charging when it's used while connected to USB as well (or at least battery indicator is stepping). The camera also has quite a wide viewing angle of 56 x 42 degrees, this is also known as field of view. It may be possible to detect pixels that are bing disguised by image processing by sweeping a thin IR source wire across the FOV and analysing the image output for deformations in the imaged line. Not something I have ever done and a purely theoretical process. There is normally no need to analyse the dead pixel content of an FPA unless an issue arises in the cameras correction process or damage is suspected, such as laser induced pixel distortion etc.

I have no doubt though that for more money there are better cameras but at this price point I have probably nothing to complain about. Most manufacturers of microbolometers clearly state the percentage of functional pixels expected to exist on a production FPA. This is generally 99.6% or 99.8% as I have already stated. Many microbolometers provide far more functional pixels as tests on the E4 and it’s dead pixel map image showed. The service mode on many cameras tells you how many pixels are marked as bad. The service modes can sometimes provide the option to carry out a fresh NUC process and create a new dead pixel map to correct pixels that have drifted badly or failed in use. Thankfully most FPA’s work their whole life with the original NUC table and dead pixel map created at the time of production. Secondly, there was a teardown from Fraser where you can see the greatly reduced quality of the inner workings. Now let’s look at what matters, image quality. The thermal sensor can produce an infrared resolution of of 256 x 192 pixels totaling 49,152 thermal pixels. This 256×192 IR resolution is really good because it is 50% more than what you get from the average thermal camera at this price level. You just won’t get that many pixels from a handheld Flir camera for this price.