-
Therm-App Noise / NETD test.
Please let me know if you have any better methods, I'd be glad to try them.
Note the temperature range is purposely kept low (1.2 Kelvin) for maximum noise. Target is a calculator in a cold room.
Therm-App just turned on:
Therm-App left on for 3 minutes:
As you can see the sensor being left to warm up for 3 minutes really made a huge difference.
Compare to FLIR E4, E40 and FLIR OneFlir E40 (hacked 320x240), Germanium (?) lens
Flir E4 (hacked 320x240), Chalcogenide Glass lens
Flir One => Lens FOL 2 (80x60)
I think the Therm-App in terms of noise and detail is somewhere between the E40 and E4. Probably closer to E40 but maybe I'm biased
The FLIR One does not produce a usable image, it's just all noise reduction artifacts.
Even though the Therm-App microbolometer (55mK typical according to ULIS) is technically worse than the E4 (~20mK) and E40 (~30mK), the lens on the E4 is very bad and cannot compare to the f/1.1 GASIR lens on the Therm-App or the f/1.3 Germanium lens on the E40. (that brings up a point... if the E40 lens is f/1.3 and achieves a way better image with a worse sensor, how bad is the lens used in the E4? It must be like f/1.8 or something...)
Opgal also sells a f/1.0 lens for the Therm-App for $250. I may purchase this in the future as it should increase sensitivity further. -
The Therm app uses a decent ULIS microbolometer combined with a decent lens. IMHO both of these are superior to the E4, hence the better low contrast image quality.
With regard to the warm up time required for best image. This is quite normal and to be expected. The microbolometer warms up as a result of its internal electronics and will stabilise at around 30 Degrees C. Until the microbolometer die naturally stabilises at its running temperature, the thermal drift offset table is unable to perform its task of compensating for die temperature changes whilst in use.
I am interested to see that the Therm App does not use a FFC shutter. It presumably uses a similar idea to the original FLIR ONE iOS that incorporated dynamic FFC. The alternative is a comprehensive FFC offset table that is created for each pixel across the temperature coverage of the camera. Such is unique to each microbolometer and created during calibration. It is an expensive technique in terms of production costs though. The quick and dirty alternative is to assume most microbolometers have pixels that drift at a similar rate and just apply a NUC table to them. Such a technique can result in individual pixels drifting outside of the predicted range but such is not always an issue in non radiometric applications.
-
nice comparison
The FLIR One does not produce a usable image, it's just all noise reduction artifacts.
Yes, there are more artefacts as real informations in the image.
https://www.eevblog.com/forum/testgear/flir-one-thermal-imaging-camera-teardown-and-hacks/msg551881/#msg551881
You know my sequence of 30 images
It's like a ghost film from hollywood. -
I am interested to see that the Therm App does not use a FFC shutter.
You may also be interested to know that the new FLIR Tau 2 cores (Tau 2.7) can operatewithout a shutterSilent Shutterless NUC™.
http://www.flir.com/cores/display/?id=54717
-
Noise and resolution test
This time with a target with slightly more contrast (enough to fill the 1.2C range).
For the resolution test I don't have hot wire so I tried to use a piece of aluminium with ridges. It didn't work nearly as well as tomas123's test. 1 square = 1 pixel of the 384x288 array.
Flir E40 (image IR_3115.jpg with 320x240 sensor pixels)
Now process the image from the Flir One ( IMG_3353.JPG ) -
Opgal Therm-App is an equal opponent for a Flir Exx
-
Given the quality of the Therm-App sensor + lens, would you reccommend it over a Flir E4 hacked to 320x240?
-
Noise and resolution test
This time with a target with slightly more contrast (enough to fill the 1.2C range).
For the resolution test I don't have hot wire so I tried to use a piece of aluminium with ridges. It didn't work nearly as well as tomas123's test. 1 square = 1 pixel of the 384x288 array.Flir E40 (image IR_3115.jpg with 320x240 sensor pixels)
Now process the image from the Flir One ( IMG_3353.JPG )
You will get a better contrasted image if you use a material with a higher emissivity, even if the aluminium was oxidised or painted it might help. Also narrow the span by increasing the bottom of the temperature span until almost all the background is gone. That will give a better image of the target. -
The reason of this terrible images is to get the maximal noise with narrow temperature span and also show the lens sharpness .
To avoid "overexposure" it's better to work with low contrast.
P.S.: please don't quote all images -
Given the quality of the Therm-App sensor + lens, would you reccommend it over a Flir E4 hacked to 320x240?
Honestly it depends... what are you using it for?
If it's electronics work etc. and you're at a desk and you would like good macro without having to mod your camera... then Therm-App definitely.
However in some situations (eg. Building inspector) the E4 is still better as it's nicer to hold and the fact that it's focus-free from 0.5m out makes it easier to use for longer range use. -
Honestly it depends... what are you using it for?
Probably a bit of electronics work. But TBH I don't really have a proper use for it. I recently got a Flir One and once the novelty wore off I realised it's quite limited in what you can do with it. I recently returned it with the intention of spending a bit more money on a decent camera.
But I guess with the i3 thermal expert coming out soon and rumors of a new seek model, I should probably wait a month or so before rushing to buy one.
-
That's exactly what I'm using it for (just pointing it at random things). In that case sensitivity and quality is far more important than actual usabilityHonestly it depends... what are you using it for?
Probably a bit of electronics work. But TBH I don't really have a proper use for it. I recently got a Flir One and once the novelty wore off I realised it's quite limited in what you can do with it. I recently returned it with the intention of spending a bit more money on a decent camera.
But I guess with the i3 thermal expert coming out soon and rumors of a new seek model, I should probably wait a month or so before rushing to buy one.
-
Got radiometric video working finally after a bit of work.
I'll post a guide later tomorrow, but basically assembling the PNGs into a .seq is easy - hard part is getting all the timestamps right. I have a little java app that does it now. -
it's funny, that the greatest user of my flir exif header reverse engineering hacks is a non Flir user
-
1. Bill Kraus has developed a SDK app that does continuous raw image saving. PM him on Flickr for a beta copy. My Therm-App -> FLIR intensity conversion is also now built into his app.
2. Convert your continuous "video" stream of PNGs into .fffs, as before. To do this in batch, you can use "forfiles" on windows.
eg.Code: [Select]forfiles /p sequence /m *.png /c "cmd /c php splitjpg.php -i flir.jpg -r @relpath -o @fname"
When you get to the step of using splitjpg.php, you need to comment out the lineCode: [Select]unlink($destimg.".fff");so that the FFFs are kept.
Upscaling and sharpening etc. in ImageMagick is not recommended as you'll make really, really big output files (768x576 @ 8.7fps for 5 minutes is over 2GB)
2.5. For correct playback speed, you need to set framerate. You can use a hex editor to edit the byte 8 bytes in front of "FocusDistance". Find where FocusDistance is using EXIFTool.Code: [Select]exiftool -v5 sequence\2015-09-27-130630-360.fff > test.txt
In the above example it has been set to 8fps (integers only).
2.75. For correct timestamps on the files themselves (which would allow for graphing), we need to edit the timestamp in the .fff files. Again, EXIFTool can be used to locate where this is.
The first 4 bytes is UNIX Time, the next 2 bytes is secondsx10^-4, the two bytes after that is even more accurate seconds (not really needed in our case), and the last two bytes are timezone.
Note: Reverse byte order
I have a small java app that will read filenames in the format YYYY-MM-DD-HHMMSS-(msec).fff (as outputted by Bill's app) and write to the fff file the correct time.See attachedcan't upload JAR, PM me for a copy. All you need to do is pass it your filename (to do for multiple files just use forfiles as before).
3. Write all your .fff files into a single sequence file. You can use window's binary copy command, eg.Code: [Select]copy /b *.fff mysequence.seq
4. Open the .fff in FLIR Tools. Beware a 7 minute file is about ~1.7GB. -
zip the jar file before uploading
-
Work in progress
Here's some before and afters of the noise reduction. As you can see radiometric data is kept completely intact.
-
You are making really great progress.
-
I was about to comment on the high noise on the picture of the sofa when I saw the temperature span is only 1.5C!
One of these days I'll pull the trigger and buy one. Only decision left: is the extra 16 fps on the 25hz model worth ~$500 to me...
-
Work in progress
I'm waiting for you code
here I asked in the E4 thread, if anybody give us a suggest, how to remove a flir logo from the Flir.Modules.Common.dll
But nobody answered me
(see attachments in the quoted post)We know, how to remove the flir logo in the E4 images.
But by using Flir Tools we get the flir logo back.
After using mediaextract (http://panzi.github.io/mediaextract/ ) I'm sure, that the logo is a PNG Byte sequence in the dll
C:\Program Files\FLIR Systems\FLIR Tools\bin\Flir.Modules.Common.dllCode: [Select]>build-win32\mediaextract.exe --formats=image "C:\Program Files\FLIR Systems\FLIR Tools\bin\Flir.Modules.Common.dll"
Extracting 0x00000000 ... 0x001a6120 (1.64871 MB) from C:\Program Files\FLIR Systems\FLIR Tools\bin\Flir.Modules.Common.dll
Writing 16.7119 kB to .\Flir.Modules.Common.dll_0017bf29.png
Extracted 1 file of 16.7119 kB size.
$ identify Flir.Modules.Common.dll_0017bf29.png
Flir.Modules.Common.dll_0017bf29.png PNG 1029x353 1029x353+0+0 8-bit sRGB 17.1KB 0.000u 0:00.000
as attachment the extracted file
Anybody knows, how to replace the embedded logo png in the dll with a transparent image of same pixel size (but smaller file size).Code: [Select]$ convert -size 1029x353 xc:none 1029x353.png
A simple binary search and replace doesn't work (after then I have the same dll size but wrong dll checksum??)
-
MJLorton did a review on the Therm-App TH quite recently (18 hours ago!):
I wonder if the TH version and non-TH are similar, the only differences I could find were spec differences: https://therm-app.zendesk.com/hc/en-us/articles/202749372-What-is-the-difference-between-Therm-App-and-Therm-App-TH-
(also I don't believe there is a topic on the TH version, which is why I'm posting it here
)
Thanks for the comparisons, by the way!You know my sequence of 30 images
It's like a ghost film from hollywood.
-
I'm waiting for you code
Still finishing up a few things, I haven't had the time.
It's literally just a nice looking interface for your PHP script and imagemagick.I was about to comment on the high noise on the picture of the sofa when I saw the temperature span is only 1.5C!
I'd personally spend the extra money on the 13mm f/1.0 lens instead.
One of these days I'll pull the trigger and buy one. Only decision left: is the extra 16 fps on the 25hz model worth ~$500 to me...I wonder if the TH version and non-TH are similar, the only differences I could find were spec differences
TH comes with the wider (but much narrower aperture) 6.8mm lens
TH is calibrated from 0 - 200C instead of 5 - 95C
TH comes with analysis software (but who cares when we can use (arguably superior) FLIR Tools)
-
Just got a quote back from Opgal regarding the 25Hz model. $1700 +$60 shipping!
After UK VAT & handling, it comes down to £800 for the base model or £1400 for the Hz. At that price I can almost get 2 base ones and do stereoscopic thermal imagingI'd personally spend the extra money on the 13mm f/1.0 lens instead.
This is starting to look like the better choice.
-
I wonder if the TH version and non-TH are similar, the only differences I could find were spec differences
TH comes with the wider (but much narrower aperture) 6.8mm lens
TH is calibrated from 0 - 200C instead of 5 - 95C
TH comes with analysis software (but who cares when we can use (arguably superior) FLIR Tools)
Okay, thanks for clearing that up!
-
Here's a rough graph of the drift of the Therm-App after startup as the sensor stabilizes. I would assume a camera with a shutter would do better in this regard.
Opgal recommends letting the camera stay on for 5 minutes before using it for Quantitative measurements.
Here's the same graph but of a single pixel as opposed to a 5 pixel average. You can see the increased noise for a single pixel.