Flir E4 Thermal imaging camera teardown

[Bud]

As i said i used VirtualDub to record my videos.
I think codecs may need to be installed separately and i used couple generic codec packages from the Net.

[Solare]

B.t.w.,

I found this line in calib.rsc:

measureInfo.temporalMk double 26.017904

As far as I understand this is actually measured temporal average noise during calibration cycle?

If so, could you guys please share which number you have in this line?

Just for comparison... it's easy - RNDIS, ftp, view file in notepad, find measureInfo.temporalMk and post the value

No need to change, download or decrypt.

Let's start the value survey

P.S:  Can I CHANGE calib.rcs without bothering CRC032 changing?

[Bud]

I do not think touching anything in calibration is a good idea. I never did.

[Solare]

@Bud,

well, maybe it wasn't that bad idea

I took that risk and altered calib.rcs file without altering the CRC. The device didn't brick!

I have progress - I have further reduced spatial pixelation (temporal I reduced in my previous session, see above), including raw noise and calibration stability by factor of 3 at least. After all, I only have this nasty horizontal lines like from humming on old TV and since all other noises were reduced and this one - not, it is now the main contributor

Well, how I did that? Naively, as always...

I discovered there is default calibration data for the Z3 camera in the folder.

I pinned my attention to the fact that double float values in actual calib.rcs files are mostly, replica from the same resource keys in the default file, but ... they are trimmed by factor of 100 in decimal accuracy. So, actual calib.rcs uses the same default values, but of smaller numerical precision. I therefore assumed that this would lead to numerical instabilities of all sorts, lower calibration accuracy (spatial) etc.

I copied the values from the default calibration files into matching keys of actual calib.rsc files and voila! I have it after reboot.

ATT.: to those who are seduced to try this!

Always backup!!!!

This will not increase sensitivity or reduce temporal noise. This only improves calibration cycles and makes it hold longer and more accurately. In my case, I have over 1 minute stability in 2C temperature span (10 seconds before this modification were enough for spatial noise start to emerge noticeably, so had to always calibrate in a dark isothermic room

As a bonus, I can now see my shadow in matte surfaces in isothermic environments... this is pretty faint signal actually, and before it was not possible to notice.

Thanks all for your neverending help, support and understanding,

I am carrying on further experimentation!

TODO: remove those fracking horizonal temporal raws noise.

[Bud]

I can now see my shadow in matte surfaces in isothermic environments...

Yikes! Spooky... Now watch ghost hunters coming in droves to buy E4s

[HighVoltage]

@HiVoiltage:
Yes, it is pretty hackable... just be aware that particular sensor in your E4 may differ a lot, as I learned... if you are lucky, you get sensor from E8, then you'll get clear, nice picture. If you are not lucky, it will be bad sensor, noisy... yes, the hack will increase resolution to 320x240, but it is questionable benefit, unless your sensor is good... this is lottery no way to know before you buy. As I said, Flir removed possibility to measure noise in 2.3.0, so you are helpless mostly in fine tuning

Hello Solare,
Thanks for that feedback. I just ordered an E4 here in Germany and should get it in the next few days.
Before I attempt any hacking, is there a way of finding out if I have the good quality, high resolution E8 sensor installed?
Thanks for any feedback

[Solare]

@HighVoltage,

High resolution is there, by design.

However, without good signal to noise ratio (low noise) this is more a toy than a tool.

After the hack will remove 'artificial noise'm you are left on your own with your particular sensor sensitivity and S|N .
If these are high and low, you are a king. If these are low and high, you are a looser

As far as I can tell there is a way to tell before actually applying hack. You'll need to install SetRNDIS.permanent.fif, reboot, open FTP client and look at the noise levels which where measured actually during calibration procedure. In 2.3.0., you only have TemporalNoise total. Good sensors will show 15 mK there, while bad sensors will approach 30.

Pretty good difference in visual comparison, especially if, as in my case, main contributor to the total is raw and column noise, which are usually smaller than pixel noise.

[PA0PBZ]

Good sensors will show 15 mK there, while bad sensors will approach 30.

calib.nrdp.ds250C_we_ap_fi_le.measureInfo.temporalMk double 14.203512   

[leoibb]

Hmmm,

just playing with my converted E4 looking out of my window at people passing by (I live on 8th floor so it is observation from 20 meters height, at the distance over 50 meters away). It was early evening and surrounding temperature was 25.

I was expecting to see walking bodies glowing white in Iron palette(assuming these are not zombies) as I believed their temperature is well above 30C. But they didn't glow white! Then, I used 'color above threshold' palette so they should appear red, while the surroundings should remain gray.

I was surprised to see that all people in my area are actually zombies, because they didn't trigger the threshold anyway... they were of course not naked, but close to that at such temperature, but they appeared only slightly warmer than surroundings, so I had to set the threshold to 25.5 before they started to emerge red in grey.

At this distance, a typical figure of man is 12 to 16 pixels high, 6 to 8 pixels wide. Quite enough I would say for successful detection. Meanwhile, my own body, although observed from 10 meters in mirror, shows 35.4 C and 36.6 with the camera at the closest possible point.

Question: are they really zombies or something is wrong with measuring distant objects temperature? 50 meters are clearly not enough for factors like humidity come into play.

P.S> I expected that even a single pixel measurement will trigger the threshold, no matter how cold the surrounding pixels are... but now I assume some interpolation or statistics is applied to a group of pixels and, basically, it results in some limiting gradient value along the chosen direction or what?

I have the same thing . How far can these see as from what I can see not too far really . Anyway of improving distance

[Solare]

Lucky...

Guys I need more statistics! Please submit as PAOPBZ did!

Report your Noise value as:

set RNDIS -> FTP -> FlashFS/System -> Calib.rsc -> calib.nrdp.ds250C_we_ap_fi_le.measureInfo.temporalMk

[Solare]

@PAOPBZ

Can I have your calib.rsc file and conf.cfg file for a closer look analysis?

You can get conf.cfg by decrypting it with Python scrips cfccfgV2.py (search on this Thread exactly match) with your SUID, which is shown on web-interface (192.168.0.2 most likely or 192.168.1.2).

cfccfgV2.py SUID conf.cfc conf.cfg in case you didn't know.

Thank you!

[alank2]

...

[TurboTom]

On my 1.19.8 F/W model, I've got a temporal noise of 14.899144 mK in the calibration file. Added 05/27/2015: F/W 1.19.8, H/W 1.1. May add row / Column noise later... Added 05/29/2015: Well, here it is, right from the internal test and calibration menu:

Temporal noise  MilliKelvin Digital Units
Pixel Noise    15.19    1.29
Row Noise    9.81    0.83
Column Noise    9.11    0.77


Spatial noise      MilliKelvin    Digital Units
Pixel Noise    9.55    0.81
Row Noise    2.58    0.22
Column Noise    4.49    0.38
Uniformity    207.72    17.66


Total noise      MilliKelvin    Digital Units
Pixel Noise    13.36    1.14
Row Noise    5.26    0.45
Column Noise    5.26    0.45


Btw, since some other owner reported they "updated" their old F/W camera accidentially, I've also done that but was able to return it back to the original configuration with some tricks... Please look up my old posts from maybe five months ago.

Cheers,
Thomas

[Solare]

Thank you all!

More reports, please!

[stefbeer]

.calib.nrdp.ds250C_we_ap_fi_le.measureInfo.temporalMk double 13.422328

[leoibb]

Can anyone tell me if there is a magnification lenses which can be clipped on for distance ?

[Solare]

I can remember only one post dedicated to telescopic system for E4... not actual, really... long ago. Unless really expensive, it will be long tube difficult to attach, which will turn image upside down

@All

Please, participate in this short survey about NoiseMk value in calibration file.

So far, I am absolute looser with mine... good idea also to add camera hw version and fw vesion to this data

Thank you!

[Fraser]

@Leoibb

What you describe is called a telescope lens and none are presently publicly available for the Ex series.

I have commercial X2 and X3 lenses and can advise that they are very expensive and do not 'clip on' to the Ex series. They cost several thousands of US Dollars new and several hundred on the secondary market.

Your challenges will be in finding appropriate lens elements at affordable prices plus making the telescope assembly and a docking ring for the E4 camera.

Aurora

[leoibb]

Thankyou for the replies. Appreciated 

[Solare]

@Aurora

I have wrote a PM for you, but there was a glitch during submitting it, so I am not sure if it was sent actually?

Thank you!

[Fraser]

Nothing received

Aurora

[donkey77]

Just received an email linking to this for anyone interested.
http://www.digikey.co.uk/en/product-highlight/f/flir/lepton-lwir-camera-modules?WT.v_sub=5339172&WT.mc_id=em_NPA1505B.UK.Send&WT.z_email=2054_NPA1505BB0UK_mainproductlearnmore--1577-FLIR&mkt_tok=3RkMMJWWfF9wsRoku6nNZKXonjHpfsX86eoqX6C%2BlMI%2F0ER3fOvrPUfGjI4DT8ZgI%2BSLDwEYGJlv6SgFS7TBMa5j2LgNWBE%3D

[OrBy]

If I am reading that right - you can get a Lepton with shutter for $131USD. Wow. I wonder if anyone is working on a portable project like Mike did to just plug one in and have a little mini imager on the cheap?

[Fraser]

We may soon see the Lepton 160x120 available via retail outlets as well. That will certainly be interesting and a great opportunity for all manner of bespoke projects 

Aurora

[mikeselectricstuff]

We may soon see the Lepton 160x120 available via retail outlets as well. That will certainly be interesting and a great opportunity for all manner of bespoke projects 

Aurora

I hope so, but suspect we may have to wait a bit for it.

If I am reading that right - you can get a Lepton with shutter for $131USD.

GBP131, not US$131

BTW the 25 deg Lepton works in the Flir One - needs some butchering of plastic though. And of course MSX doesn't work. Just did it as a quick hack to test it as it was easier than taking apart my tiny cam...