Flir E4 Thermal imaging camera teardown

[mikeselectricstuff]

K.V. - yes, it looks like your E4 is still an E4. For example, here is the embedded thermal image from the African Gray parrot file:

The easy way to tell is the presence of the additional measurement modes and PiP display mode. The usual cause is getting the CRC01 wrong somehow

[amyk]

mrflibble, if you could do that analysis on Mike's sensor it would be interesting to see the results - he posted the dead pixel map a while ago and everyone commented on the huge number of duds (~200?). But now we're reasonably certain that there's some sort of binning going on.


Is there something interesting about the thermal profile of birds? There's been an unusually large number of pictures of them so far...

[mrflibble]

mrflibble, if you could do that analysis on Mike's sensor it would be interesting to see the results - he posted the dead pixel map a while ago and everyone commented on the huge number of duds (~200?). But now we're reasonably certain that there's some sort of binning going on.

Do you happen to have the link? This thread is so bleeping long... Also, did he post the .gan file? All I recall is him showing a snapshot with rset .image.flow.maps.combGainDeadMap.pixReplace false. While informative, that is not the same as the .gan file.

Maybe you are thinking of this one?

[Fraser]

I have noted comment on the number of dead pixels in a particular micro-bolometer.

Before anyone gets overexcited about dead pixels, please consider that the detector is not a common CCD and is prone to dead pixels. It goes with the territory, especially when working on a low budget. The threshold for a FLIR provided micro-bolometer for ANY consumer grade camera is 99.9% operational pixels. Some cheaper grades of micro-bolometer are 99.8% operational.

The maths says:

320x240 = 76800 pixels

99.9% = 76800 * 0.1% = 76.80 dead pixels
99.8% = 76800 * 0.2% = 153.60 dead pixels

It is my belief that the micro-bolometer fitted in the Ex series is one of the cheaper 99.8% operational sensors. Note the operational percentage applies to a model of sensor and not 'binning' within a model of sensor production. 99.8% is a perfectly respectable micro-bolometer and that is why a thermal camera has dead pixel processing. Even my $50,000 thermal camera has dead pixels. For newbies the term dead pixel may appear scary , but please consider this normal for the technology, the price point of the camera, and the amazing value for money when compared to cameras like my PM695.

I note comments regarding the quality of the lens used in the Ex series. Again I must point out that the thermal imaging industry normally uses very expensive Germanium lenses that cost more than the retail price of the complete E4 ! Efforts were made to reduce the cost for consumer cameras. The need for a huge depth of fields and fixed focus lead to a small aperture simple 'reverse telescope' lens system. The material from which the lens is made is not pure Germanium, it is Chalcogenide Glass. This material has made moulded lenses possible with a substantial reduction is manufacturing costs. It is comparable in performance to a Germanium lens but not equal. The choice made by FLIR was to provide a decent, if not perfect, Chalcogenide Glass lens as opposed to a large Germanium lens as fitted to the Exx series. That as part of the cost reduction process. If you want a quality 30mm Germanium multi element focussed lens system, expect to pay around $2000 to $5000 for just the lenses, depending on quality.

For newbies to thermal imaging, remember the purpose of the product and consider the true resolution of the system. The resolution is that of a common cheap low end web cam, and not that of a DSLR. An amazing quality lens structure is NOT required, hence the common use of simple single or limited element count lens structures, even in the industrial cameras. These things are not designed to produce 'artistic' images with zero barrel distortion etc 

Sadly, if anyone wants really high quality micro-bolometers and lenses guaranteed, they will need to open their cheque book and expect to pay the going rate, which last time I bought such a camera was GBP35000 ($52500). That was two years ago. When this is considered, I think many will agree that $1000 for a 320x240 camera with a less than perfect micro-bolometer and  lens is still quite a bargain.

Newbies to thermal imaging.....Enjoy your cameras and stop worrying about dead pixels and lens perfection 

To those who are investigating such issues, I know that this is for your own enjoyment and interest and I enjoy reading your posts. I just don't want newbies thinking that your excellent findings mean the E4 is cr*p as it most definitely is not, especially when in E8+ guise  Having the manual span mode has been a godsend in my recent use of the camera. Thanks to all that have worked to improve on the base configuration and released the cameras full potential.

[London Lad]

A question for you guys using a close up lens in a holder:-

I've just received my ZnSe lens to go in one of the holders designed by Georges80 and note that it has a flat and a convex side.
 
Did you guys mount yours with the convex side facing the camera lens or away from it? Mine seems to work both ways???

[Fraser]

Flat side towards target, convex towards camera

To see why, look here:

http://www.thorlabs.de/newgrouppage9.cfm?objectgroup_id=1780&pn=LA9410-F

The light path is in reverse to that shown in the picture. In very close focus situations plano-convex are common but users often operate them with the convex towards the target a it 'looks right'. A mistake but the lens does work in reverse, just not as well.

My inframetrics meniscus close-up lens in configured in this way with the concave side facing the target. It does look odd but it is how it is supposed to me.  The longer focal lengths from China are bi-convex but my 50mm FL lens is also plano-convex like yours.

[London Lad]

Flat side towards target, convex towards camera

Thank you!

[mrflibble]

I have noted comment on the number of dead pixels in a particular micro-bolometer.

Before anyone gets overexcited about dead pixels, please consider that the detector is not a common CCD and is prone to dead pixels. It goes with the territory, especially when working on a low budget. The threshold for a FLIR provided micro-bolometer for ANY consumer grade camera is 99.9% operational pixels. Some cheaper grades of micro-bolometer are 99.8% operational.

Indeed. Especially when considering that what we paid for is for a 80x60 sensor. Even for example Mike's sensor which has a reasonably large number of bad pixels is perfectly capable of creating good quality thermal images.

Newbies to thermal imaging.....Enjoy your cameras and stop worrying about dead pixels and lens perfection 

To those who are investigating such issues, I know that this is for your own enjoyment and interest and I enjoy reading your posts. I just don't want newbies thinking that your excellent findings mean the E4 is cr*p as it most definitely is not, especially when in E8+ guise  Having the manual span mode has been a godsend in my recent use of the camera. Thanks to all that have worked to improve on the base configuration and released the cameras full potential.

Running those FFTs was basically along the lines of "lets see what we can see". For example I wasn't really looking for any high frequency components in the spectrum. But hey, the numbers are the numbers. So after that I did that bit of seperate statistics for odd/even rows/colums, and hey, that confirms it. Nothing ground breaking or worrisome, just a bit of numerical fun.  And the exact same goes for bad pixel count. Besides, this is a perfectly good excuse to practice using numpy/scipy.

[Fraser]

Mrflibble,

I love the work you have done, and are doing....I don't claim to understand it all, but that is why it is interesting. You and I know how good the E4 is in use, but I didn't want new arrivals to the thread to think the E4 was a flawed product.

Please keep investigating the Ex secrets...it makes very interesting reading.

[cynfab]

bernroth, could you do a "cat /proc/sys/net/ipv4/conf/all/arp_accept"

arp_accept is 0 on my system.
I think they might have fixed the whole RNDIS issue somewhere between 3.8 and 3.11

If 3.11 available on your system?

Try:
aptitude search linux-generic-lts-saucy

If not, well, I would recommend to use Mint 13 as it's supported till april 2017 and has always all new kernels from the newest ubuntu distributions available.

bernroth,

It's not a kernel issue.
Just got some to time to poke at this again, what led me to do the arp_accept = 1 is that on all of my linux boxes it was set to 0 and when the camera sent it's GARP packet, none of my arp cashes got updated. I mistakenly thought that was a problem. After you & mrfibble reported success WITHOUT arp_accept = 1, I tried just changing my static IP to 192.168.250.1 and it "just worked(tm)". Tested on Mint 14's 3.5.0-17-generic and Ubuntu 9.10's 2.6.31-23-generic.
 
Win 7 and the camera go through quite a dance when the camera is plugged in, first starting at 192.168.250.x then switching to 192.168.0.x or 192.168.1.x depending on what comes back from various ARP and PING requests.

When we had tried a dynamic IP, the linux dhclient didn't do anything but a DHCP_DISCOVER which the camera didn't respond to, then dhclient kept retrying till it timed out.

When we did a static IP of 192.168.1.1 or 192.168.0.1, the camera was at 192.168.250.2 and waiting. The camera would respond to an DHCP_REQUEST, but I could never get that to work by modifing dhclient.conf. Probably a mistake on my part in trying to match the DHCP_REQUEST format that the Win 7 box sent.

With a static IP of 192.168.250.1 the camera was happy at 192.168.250.2, and would talk telnet/ftp/http.

[tomas123]

The material from which the lens is made is not pure Germanium, it is Chalcogenide Glass. This material has made moulded lenses possible with a substantial reduction is manufacturing costs. It is comparable in performance to a Germanium lens but not equal.

here is a compare of a Flir E4 (FOL 7) with fixed focus Chalcogenide Glass lens and a Flir E40 with a Germanium (?) lens (FOL 18)
both cameras enhanced to 320x240

this is a coffee machine
Distance Flir E4 ca. 110cm (Datasheet 45° field of view)
Distance Flir E40 ca. 190cm (Datasheet 25° field of view) 

NETD E4 vs. E40
https://www.eevblog.com/forum/testgear/flir-e4-thermal-imaging-camera-teardown/msg342873/#msg342873

all images converted with exiftool and imagemagick
E4 use a png with reverse byte order - see code for changing the byte order

Code: [Select]

exiftool -b -RawThermalImage FLIR0238.jpg | convert - gray:- | convert -depth 16 -endian msb -size 320x240 gray:- -auto-level -resize 800x -depth 8 e4-gray.png
convert e4-gray.png iron.png -clut e4-iron.png

exiftool -b -RawThermalImage IR_0480.jpg | convert - -auto-level -resize 800x -depth 8 e40-gray.png
convert e40-gray.png iron.png -clut e40-iron.png

download color palettes from this post
https://www.eevblog.com/forum/testgear/flir-e4-thermal-imaging-camera-teardown/msg342072/#msg342072

[Fraser]

The Exx series has an undoubtably better focus. No argument. But this is comparing a fixed focus lens with a manual focus lens. Manual focus will win out as it does not have the compromises of the fixed focus optics. In this case, even a Germanium lens would not fix the image softness. I suppose this is why all of my industrial cameras are manual or auto focus. I do have to do a lot of re-focussing whilst using them though. The DOF on my PM570 and PM695 cameras is very shallow. The fixed focus lens allows FLIR to reduce cost whilst making life easier for the user. All compromises that effect the final image. but hopefully not to the extent of making the camera a poor performer.

[mrflibble]

When we had tried a dynamic IP, the linux dhclient didn't do anything but a DHCP_DISCOVER which the camera didn't respond to, then dhclient kept retrying till it timed out.

When we did a static IP of 192.168.1.1 or 192.168.0.1, the camera was at 192.168.250.2 and waiting. The camera would respond to an DHCP_REQUEST, but I could never get that to work by modifing dhclient.conf. Probably a mistake on my part in trying to match the DHCP_REQUEST format that the Win 7 box sent.

Yup, I also noticed the lack of response to the DHCP_DISCOVER. But that's not the root problem either. Because when I noticed the lack of DHCP_DISCOVER response I thought I'd be clever and fake an existing dhcp lease, so that dhclient goes straight to the DHCP_REQUEST stage. When I wireshark it the request looks exactly the same as for win7. Buuuut, no reply there either. So something else is going on. But at that point I went screw it, I will just use a static config. Neeeeeeext.

With a static IP of 192.168.250.1 the camera was happy at 192.168.250.2, and would talk telnet/ftp/http.

Yup. That's my setup as well. With allow-hotplug you don't even have to ifup the interface. Just plug it in and all good to go.

[tomas123]

@Aurora

I have the images not rated.

The Flir E4 -> E8+ has a great cost-benefit ratio.
The price of a original E8 is overpricing opposite a E60 (both have native 320x240 resolution)

[okent]

Two more successfully converted today.  version 1.19.x
Thanks to Mike and everyone involved in this awesome hack!

[Rufus]

The print view of this topic now renders a blank page, guess it got too big for SMF as well.  Shame. Apart from easy in browser searching the print view is also useful for saving a local copy.

[Fraser]

@tomas123,

Apologies, I misunderstood the reason for your comparison of image quality. I totally agree that the E8 is a very poor match to the Exx series and offers very poor value in comparison at that retail price point.

It was interesting to see the lack of fine detail in the Ex image when compared to the Exx image. As I stated, not unexpected but a very useful comparison all the same.  The E40 gives a decent image 

When using a close-up lens there may be an improvement in detail within the image if the auxiliary lens effectively becomes a focussing lens. I shall have to experiment.

For the mechanically minded it would be feasible to fit a much better lens to the E4, in place of the small OEM fitment, but the calibration would suffer and a decent primary lens plus variable focus assembly would need to be sourced. They sometimes appear at reasonable prices as ex military equipment in the USA.

[mikeselectricstuff]

@tomas123,

Apologies, I misunderstood the reason for your comparison of image quality. I totally agree that the E8 is a very poor match to the Exx series and offers very poor value in comparison at that retail price point.

It was interesting to see the lack of fine detail in the Ex image when compared to the Exx image. As I stated, not unexpected but a very useful comparison all the same.  The E40 gives a decent image 

When using a close-up lens there may be an improvement in detail within the image if the auxiliary lens effectively becomes a focussing lens. I shall have to experiment.

For the mechanically minded it would be feasible to fit a much better lens to the E4, in place of the small OEM fitment, but the calibration would suffer and a decent primary lens plus variable focus assembly would need to be sourced. They sometimes appear at reasonable prices as ex military equipment in the USA.

The  construction would significantly limit how big a lens you could get in there before you have to start chopping into the magnesium frame. I don't think the thread is any kind of standard that I could find - I did compare it to a few cable gland threads but didn't find a match

[Fraser]

Agreed. The fitting of a different lens would involve significant changes to the E4 lens assembly. I would not expect to find anything that would be a simple fitment. I was thinking of the complete lens assemblies that are sometimes available Ex military. These would have to be grafted into the E4 and some serious modifications undertaken. The user would effectively be re-engineering everything forward of the micro-bolometer. 

Who would bother? Well if someone wanted to turn the E4 into a long range surveillance thermal camera or just better optical  performance is needed without accurate temperature read-out, such a project would make financial sense as opposed to buying one of the Exx series or buying an Exx plus a telephoto lens. This kind of project would use the E4 as a cheap building block for a more advanced camera. At $1000 it is still a bargain 320x240 thermal core when compared to the FLIR Tau 

You are right to point out that this would not be a trivial operation though 

I have an Agema 880 with a very high quality 12 Degrees Germanium lens that measures approx. 70mm  diameter.  I have all the required germanium optics in the 880 lens mount (removable) to graft it onto virtually any camera core. That is a future project I have planned for upgrading an ISG Talisman Wasp BST 320x240 fire fighter TIC. It has a advantage of a X2 digital zoom    The 880 is an expensive museum piece and needs liquid nitrogen that I cannot easily source, so it is basically a paper weight at the moment.

I personally will not hardware modify my E4 as it does all that I need of it. I can fall back on my PM570's and 695's for detail work, and I have the required auxiliary lenses for this already.

Thermal imaging can be a very interesting alternative to conventional digital photography and I enjoy experimenting with the technology, but I do not have the expertise in the software improvement that you and others are demonstrating here. I am very much a hardware tech when it comes to TIC upgrades 

Axis do a range of CCTV thermal lenses that might be an option for an E4 lens upgrade. Picture attached. I saw a 19mm AXIS lens for $250 recently on e*ay USA.

[okent]

Once you get into engineering a retrofit optic I think you would be getting close to the cost of the FLIR Scout PS32.
Same core and refresh rate and about 3K in cost.
Granted, I would love to have an optic I could attach to the front of my E4 to allow longer range scanning but as you stated I just don't think it is reasonably possible.

[olsenn]

I was actually surprised that the Scout series is also only 9Hz refresh rate. I figured since it was designed for hunting (primarily) it would be faster to accommodate for tracking a moving animal. However, I suppose that pesky export license just wouldn't be worth the hassle.

[osteron]

Received my E4 in germany:

Calibration Date  03 Dec
Serial Number   6390 89xx
Firmware 1.19.8
Model E4 1.1

[Fraser]

E4 1.1 ?

That's a new HW version. My E4 is a 1.0

Could you please provide the camera module version information.

To access this got 'Settings' in the menu. Select 'Device Settings', select 'Camera Information'.
Now press the right side of the navigation pad for 10 s3econds or more. Another menu will appear.
Select 'Version information'

Grateful if you will detail the list the module versions. You can truncate the name of the module for speed if that helps as there are a few modules to list.

To exit the menu use the left navigation pad button.

Many thanks

[ixfd64]

@osteron

Did the hack still work?

[Fraser]

Not wanting to be mean or untrusting but we have already had a hoax first poster with a fictitious firmware version. Until I see a screen picture and the module versions I will treat this report as a typo and unconfirmed. If another E4 is found with the 1.1 HW we will have corroboration of the report.

The next question will be, does the unit accept the improvements ?. If not it will be a case of working out what has changed.