Flir One Thermal imaging camera teardown and hacks

[tomas123]

as thread holder I collect here some infos:

the new Flir One 2. Generation (Flir1G2) has a Lepton sensor 160x120, but make cropped images (120x90) by using the Flir apps

you can see it with the simulator of Flir SDK app

The sdk app has crop factor switch.
If you unplug the Flir One and then start the SDK.app then all simulation images are uncropped.
If you plug in the Flir One then the simulation get the config values from the real camera and simulation images are cropped!!

see attachments (images from SDK simulation):
- both images from simulator loaded in FlirTools.app to shrink the file size
- very crazy: FlirTools for Android resize the image to the crop factor
   

see the difference in image size

... the FlirOne app makes only cropped images

The Flir1G2 give us a cropped image of RAW sensor!!

You can simple test it with the Flir One SDK. There is a sample app (compile it with Android Studio) which show you the real sensor image without fake noise etc.
select "Thermal Linear Flux 14 Bit Image"

As attachment two images from my notebook:

(1) a flir radiometric image with the FlirOne.app (= SDK)


(2) and at the same position two screenshots from my android device


Look at the keyboard and you see the difference in field of view (FOV) and thermal resolution!!

follow the links for more informations

the simulator has raw images for simulation

an interesting detail:

Inside the android sdk app (see post above) there is an simulator which works without a real FlirOne.
http://www.file-upload.net/download-11032220/FlirOneSDK.apk.html
The images for the simulator are real Lepton Sensor (raw) USB frames on which you can switch the view between large FOV "Thermal Linear Flux 14 Bit Image" and the cropped Flir "Thermal Image"

You can import the images with PS / Open as ... "Photoshop Raw *.RAW"

see PS screenshot (it's a rotated coffee cup)

If you test the SDK with Android Studio, try also this:

Here you can download a working SDK sample with the extracted java.class (see my post above)
 see path FLIROneSDKBundle\FLIRONEExampleApplication\app\src\main\java\com\flir\flironesdk\usb

have fun
http://www.file-upload.net/download-11040066/FLIROneSDKBundle04-Basis.zip.html

[cynfab]

Other than the change in the .so library, there doesn't seem to be too many differences in the classes.jar code.
BUT... when I recompiled with the new .so and a freshly installed Android Studio 1.5 preview, the app runs on both the tablets that it crashed on before.
This is good.
Tried the your decompiled classes and they compile and run just like the app with classes.jar.  ;>))

[tomas123]

 
I'm sure that the crop factor is only a software switch, because the SDK simulator use full size raw values and also shows cropped images after loading the configuration from the real Flir One

[cynfab]

With the example app and the decompiled java classes as a base we should be able to build an app that provides true 160 x 120 Lepton 3 quality images. With adjustable emissivity.
I've been impressed with the detail in the images from the example app, and how seldom the "tuning" happens. especially vs. the Seek.
ImageJ handles the raw Lepton images. ImageMagik should as well, but I wasn't able to get it to work. Need to fiddle with it some more, I'm sure that exiftool etc. has a solution.

[tomas123]

ImageJ handles the raw Lepton images. ImageMagik should as well, but I wasn't able to get it to work.

Code: [Select]

$ convert -depth 16 -size 164x120 gray:00001-lep 00001-lep.png

All image calculations (palette, MSX etc) be happen in software with libjnidevicewrapper.so.
This is also the reason of lags with weak android hardware.

The USB frames are only raw Lepton frames and a visual camera frames.

[tomas123]

here is a real live sample (shot after a small warm up time of about 2 minutes):

It’s the same shot. Compare the different FOV (field of view)

I saved with the SDK.app simultaneously a upscaled Flir Radiometric JPG  and a real Lepton ThermalLinearFlux14BitImage.

Afterwards I rebuild with my old panorama script (see my footer) a real size 160x120 Lepton radiometric jpg (a Flir format).
You can load this sample jpg images in Flir Tools and compare the quality.

First a original image shot with the Flir App.
The App crop    the Lepton sensor to about 120x90 Pixel.
Please note the artefacts/patterns!
Flir makes a nice lens distortion correction of the Lepton sensor for best MSX overlaying 




real  Lepton sensor 160x120 (no image postprocessing and with noise/grain because the temperature spread is only 4 Kelvin)

[tomas123]

If you disappointed about the grain of the Lepton image above (with 4 Kelvin scale) look at this images from a motherboard (27 Kelvin spread) shot with Flir1G2
Remarkable is the really short integration time. The rotating fan is sharp!

image shot from FlirOne App (about 120x90 Sensor Pixel)


with the modified SDK App I shot a full size Lepton image (without image postprocessing): 160x120 ThermalLinearFlux14BitImage


a good visible barrel distortion

as attachment the radiometric jpgs for using with flir tools

[OrBy]

Remarkable is the really short integration time. The rotating fan is sharp!

Give that we have seen 30Hz Lepton's out there I don't think it's unreasonable that it may be operating at 30hz internally and outputting 9Hz for ITAR reasons. (it was mentioned that the the E4 does this)

a good visible barrel distortion

Thinking about it - this is likely the reason for the cropping. I would think it is a lot easier to MSX with a distortion free thermal image then try to "un-action cam" it then MSX it on the fly.

Afterwards I rebuild with my old panorama script (see my footer) a real size 160x120 Lepton radiometric jpg (a Flir format).

Is there anything that that script cant do? 


Now that its been discovered that that the effective thermal resolution of the G2 is only 120x90 for the end user and not the advertised four times (160x120) of the original 80x60 G1 - how long before we see a class action lawsuit? 

[tomas123]

Now that its been discovered that that the effective thermal resolution of the G2 is only 120x90 for the end user and not the advertised four times (160x120) of the original 80x60 G1 - how long before we see a class action lawsuit? 

I'm sure, that the FlirOne G1 is also cropped.
So it's 4x a true statement 

after the sensitivity tests above it's time to make a resolution test
...
one square = one sensor pixel of Flir One G1

....

compare it with the lepton 3 image of motherboard (see 2 post above)

Thinking about it - this is likely the reason for the cropping. I would think it is a lot easier to MSX with a distortion free thermal image then try to "un-action cam" it then MSX it on the fly.

Hm, the crop factor is to large for a good faith.
Combined with the mystic pattern generator (same algorithm like in Flir1G1) is this a  noticeable degradation of the Lepton sensor.

Give that we have seen 30Hz Lepton's out there

Integration time (like shutter speed) and frame rate are different things. You can't make a sharp photo from a rotating fan with 1/30s.
see here for a sample: http://www.flir.co.uk/cs/display/?id=41923

[encryptededdy]

AFAIK integration time isn't really a thing with microbolometers. It's mainly for photon counting (cooled) sensors.

Remember, a microbolometer is an array of thermistors - so it doesn't matter how fast you read it out, you will always be limited by the thermal time constant of the microbolometer. This is why most 9Hz cameras read at 30Hz and either drop frame or merge frames - you can't have a slower "shutter speed / integration time" and get better noise performance. All you can do is merge frames and reduce readout noise.

This is also why to read higher temperatures microbolometers need a high temp / ND filter, whereas a cooled camera can just reduce integration time. The microbolometer's thermistors are always sensing, and we can't change that.

I'm not an expert here so perhaps someone like Fraser can confirm, but that's my basic understanding. I found it out when I was getting really confused why cameras were reading out at 60Hz and dropping frames, and why they can't just use a faster "Shutter speed" for higher temperatures.

edit; a quote from FLIR

The thermal time constant should not be mixed up with "integration time" which is a property of photon detectors.
A microbolometer detector is constantly open to incoming radiation, and responds to changes with the time constant.
A photon detector collects energy only during the integration time which is a parameter that is set by calibration.

[tomas123]

Thanks for clarification 

[tomas123]

I understand your question but I neeed some more informations:
Which software do you use for reading the sensor images with "better resolution"?

In this thread we discuss about the low level frame transfer

Code: [Select]

{"type":"setOption","data":{"option":"autoFFC","value":true}}
{"type":"setOptionStatus","data":{"option":"autoFFC","value":-1}}


see for "What is FFC?":
http://www.flir.com/cvs/cores/knowledgebase/index.cfm?CFTREEITEMKEY=327&view=35774


Please post a photo from your lens hack!!

[cynfab]

Using the FlirOne SDK, you can perform manual tuning, or enable/disable automatic tuning.

    performTuning()
Call this method if tuning is required or requested for high thermal accuracy

    setAutomaticTuning(boolean shouldAutomaticallyTune)
Disable or enable the automatic tuning feature of the device.

As well as querying the device tuning state.

[Thermo Andy]

My Flir to Cam Adapter and my Tablet With Tune Automatic....

[tomas123]

Wow, a Flir2LAN adapter with a video camera
In Germany we say: "Dem Ingeniör ist nichts zu schwör" - "nothing is too hard for an engineer"

[Scottjd]

I now have the flir 1.0.9 firmware update pkg.
Lets see what that contains 

now i just need to find a way to depkg the file

I wonder what firmware 1.0.11 has now that it was released for the version 1 Flir One.

I just picked one up for cheap since version 1 is coming down in price and the V2 seem to be out of stock.
I don't have an iPhone 5/5S, but I did get an iPod Touch 6th Generation 32GB as part of the fluke tools promotion. They didn't have anything else in the promotion I wanted, so I was going to sell it.
But I realized the 6th gen iPod touch (current version) had the same processes as the iPhone 5! I think it's the 64 bit A8 with the same mobile M8 co-processor. So I figured I would save some money and buy the first gen Flir One. It works great.
A little heavy card stock slip of cardboard behind the iPod touch and it even fits in the case well.
Both seem to have the same rate of battery drain when doing Flir videos, so it's a nice match.

I also noticed the app shows a higher max read, and the hardware version is "L", not sure if that makes a difference. But the first version is supposed to have a cap at 218F I think and it showed greater then 248F like on the second version Flir one.

I updated the firmware to the latest, it came with 1.0.8. I had to do an update to 1.0.9 then again another update and it jumped to 1.0.11. On the 08 firmware I wasn't able to see any studs in my walls. After the 11 update I went around the house and noticed it was seeing the wall studs.

Here is the video, call it a thermal tour of my house 
https://youtu.be/17TCYtUv67Q

[tomas123]

interesting, in November 2014 it doesn't work with an Ipod touch 5th Generation

the Ipod touch 5th generation fits great in the Flir One hardcase

but the flir one app doesn't run

[Scottjd]

interesting, in November 2014 it doesn't work with an Ipod touch 5th Generation

the Ipod touch 5th generation fits great in the Flir One hardcase

but the flir one app doesn't run

I know, I saw that post also. I took a chance figuring I can always mod the unit to use on my iPad. Theirs a few videos on YouTube showing it modded for the iPad and iPhone 6/6S or a extention cable on eBay. I don't know if the cable works, I didn't try it.

But my research also showed that the iPod touch 6 used the same processor as the iPhone 6, the A8 processor. Since it was designed to run with the 5/5S that has the A7 processor I thought it might work.
The iPod touch 5th gen is powered by the A5 processor, which I suspect the A5 doesn't have enough power to run the Flir one 1G. So the app is probably checking to see what processor the device is running.

Also this A5 in the fifth gen iPod touch is a 32 bit processor. The A7 was the first of the 64 bit processors from apple and it made a big jump in the performance.

I also think the iPod 5th had 512MG RAM and the 6th has 1GB of RAM.

[Scottjd]

I also moticed the Flir One V1 says it only goes to 212F/100C but in several times if seen it display higher.
Maybe this is from the app/firmware Updates?

I also out a little video showing it is running with the iPod 6th gen. Some people suspected I sit jail broke an iPhone to say iPod so I figure a video proves all.
https://youtu.be/kVE4_fZcobc

The attached picture is of my 02 sensor in my exhaust, one for the few things in the house that I own and will heat up past 100C.

[tomas123]

I made a Flir One G2 run down test as answer to this seek post:

Drifting of 8 random pixels in 1 hour. On some pixels temperature readings go up and on some readings go down.
As you can see thermal pixels need at least 20min to output stable readings.



Explanation of values:
Dark red line is a pixel which starts with value 7537 and ends with 6818. That is 719 less.

with the source code from
https://www.eevblog.com/forum/thermal-imaging/question-about-flir-one-for-android/msg846130/#msg846130
I wrote all logs to file and generated a excel sheet.

This was a cold start of the Flir One and the log begins 1 minute after shutting on.
The battery reaches only 57 minutes, but it's more as expected:

The FLIR One houses its own battery source that can power the device for more than 45 minutes of continuous use.

The blue line is the average temperature of 2x2=4 center pixels. The measured temperature was a wall with a constant temperature over the hour.
The F1 G2 needs 10 minutes until stabilization.
The vertical green lines mark the FFC shutter movements.

The shutter interval is constant 135 seconds after 5 minutes runtime.

After FFC the temperature needs 30 seconds to settle (2 sec down + 10 sec up + 20 seconds to stabilize)   
a sample at 00:25:50

Code: [Select]

Time    Spot FFC
0:25:50 19,6
0:25:51 19,6 Shutter
0:25:52 19,3 Shutter
0:25:53 19,1
0:25:54 19,0
0:25:55 19,1
0:25:56 19,3
0:25:57 19,5
0:25:58 19,7
0:25:59 19,9
0:26:00 20,1
0:26:01 20,2
0:26:02 20,3
0:26:03 20,3
0:26:04 20,3
0:26:05 20,3
0:26:06 20,3
0:26:07 20,3
0:26:08 20,3
0:26:09 20,2
0:26:10 20,2
0:26:11 20,1
0:26:12 20,0
0:26:13 20,0
0:26:14 19,9
0:26:15 19,9
0:26:16 19,8
0:26:17 19,7
0:26:18 19,7
0:26:19 19,7
0:26:20 19,7
0:26:21 19,6
0:26:22 19,6
0:26:23 19,6
0:26:24 19,6
0:26:25 19,6
0:26:26 19,6
0:26:27 19,5
0:26:28 19,5
0:26:29 19,5
0:26:30 19,4
0:26:31 19,4
0:26:32 19,4
0:26:33 19,4
0:26:34 19,4
0:26:35 19,4
0:26:36 19,4
0:26:37 19,4
0:26:38 19,4
0:26:39 19,5
0:26:40 19,5
0:26:41 19,5
0:26:42 19,4
0:26:43 19,5
0:26:44 19,4
0:26:45 19,4
0:26:46 19,3
0:26:47 19,3
0:26:48 19,5
0:26:49 19,4
0:26:50 19,5
0:26:51 19,5
0:26:52 19,4
0:26:53 19,4
0:26:54 19,5
0:26:55 19,4
0:26:56 19,4
0:26:57 19,5
0:26:58 19,5
0:26:59 19,5
0:27:00 19,5
0:27:01 19,4
0:27:02 19,4
0:27:03 19,5
0:27:04 19,5
0:27:05 19,5
0:27:06 19,5
0:27:07 19,5
0:27:08 19,6
0:27:09 19,6
0:27:10 19,6
0:27:11 19,6
0:27:12 19,7
0:27:13 19,7
0:27:14 19,6
0:27:15 19,6
0:27:16 19,6
0:27:17 19,7
0:27:18 19,6
0:27:19 19,6
0:27:20 19,6
0:27:21 19,6
0:27:22 19,7
0:27:23 19,7
0:27:24 19,7
0:27:25 19,7
0:27:26 19,7
0:27:27 19,7
0:27:28 19,6
0:27:29 19,6
0:27:30 19,6
0:27:31 19,8
0:27:32 19,7
0:27:33 19,7
0:27:34 19,7
0:27:35 19,7
0:27:36 19,7
0:27:37 19,6
0:27:38 19,6
0:27:39 19,6
0:27:40 19,5
0:27:41 19,6
0:27:42 19,6
0:27:43 19,6
0:27:44 19,7
0:27:45 19,7
0:27:46 19,6
0:27:47 19,6
0:27:48 19,6
0:27:49 19,6
0:27:50 19,6
0:27:51 19,6
0:27:52 19,6
0:27:53 19,6
0:27:54 19,6
0:27:55 19,7
0:27:56 19,7
0:27:57 19,6
0:27:58 19,7
0:27:59 19,6
0:28:00 19,6
0:28:01 19,7
0:28:02 19,7
0:28:03 19,7
0:28:04 19,7
0:28:05 19,7
0:28:06 19,7
0:28:07 19,6 Shutter
0:28:08 19,4 Shutter
0:28:09 19,0
0:28:10 18,9
the corresponding battery diagram: 


PS: all described effects (shutter interval, temperature drift) you also get with the Flir One Ap from google play store

[sharpharp]

Hi,

If I buy a new Flir e4, what is the current firmware they come with?

Can I do the downgrade to run the hack?

I read that the resolution can be upped to an e8, but what about the thermal sensitivity?
On the e4 it is <0.15, on the E6 onwards its <0.06 -Will the hack also unlock the thermal sensitivity?

Is there any other reason why to buy an E6 or E8 which can't be achieved by this thread?

Also, a main dealer in the Uk told me that the E4s come with the latest firmware which cannot be modified, is this correct?

Thanks

[-jeffB]

You'll have better luck if you ask in the E4 discussion thread. Better yet, spend some time browsing the last month or two of activity on that thread; it should answer your question about being able to downgrade and do the hack (yes, you can).

[mangler]

So,

It seems pretty obvious the Flir One G2 has a very good sensor, somewhat crippled in firmware, but not too bad. The battery seems to be a big issue, not very long lasting, or user replaceable. Makes the whole thing disposable, which I am guessing is the point for Flir.

My question:

With all the 3d printers out there, has anyone designed a better housing for this nice sensor, with appropriate replaceable battery? I am thinking something along the line of the therm-app pistol grip design, with a cell phone battery?

[in.Vitro]

....

another question is: has anyone stick a better lens in front of the sensor to get much better results?

[tomas123]

the lepton 3 lens in F1G2 is good enough for the resolution 160x120 but has some strong distortions

I haven't yet repeated my heating wire test with a F1G2, but the screen of an uncropped SDK app looks good
see https://www.eevblog.com/forum/thermal-imaging/flir-one-thermal-imaging-camera-teardown-and-hacks/msg551882/#msg551882

with the modified SDK App I shot a full size Lepton image (without image postprocessing) 160x120


and the (rotated) center of the image with marked pixels