Flir E4 Thermal imaging camera teardown

[mamalala]

]The problem is that (if they have any sense) the pinout will be adapted to suit the PCB of the product - even a single pin swap would mean a very different bitstream. Add to that this product has the visible camera stuff handled by the FPGA.

It would however be interesting to see what signals & voltages are used to drive the sensor. It's a bit tricky to get at though - would need to make a riser board.

That's something i would not be so sure of. I have seen different products from the same manufacturer re-using circuit blocks. Sure, the actual layout was sometimes different, but the interconnections remained the same. of course none of that used FPGA's, but microcontrollers instead, but even those could have benefited from a slightly different pin usage depending on the layout. But instead they reused the same block, with the same interconnections. Of course i am just specualting here that FLIR does the same, so it may well be that i am completely wrong here.

Greetings,

Chris

[mamalala]

Edit: That reminds me about something that i was wondering for quite some time: is there anything like a disassembler for FPGA bitstreams? Maybe not something as sophisticated as IDA, but still? Or is the IP about what goes on inside an FPGA such a "mystery" that there is simply no way to produce such a tool?

There were a couple projects trying to reverse-engineer bitstream format for Xilinx FPGAs. It was shut down by Xilinx lawyers almost immediately as it became public. It is still possible to find a PDF describing the effort, but referenced files are all gone from public view.

Sadly, this doesn't surprise me in the least ;(

Greetings,

Chris

[mikeselectricstuff]

I had a play with the TargetNoiseMk setting and it does appear to be added noise.
There is a small, but noticeable difference between default value of 135 and a low setting of 13
I can only assume it's to reduce banding and/or give an impression of sharpness
(These pics are with MSX, but low light so mostly thermal)

[M. András]

thats an absolute joke. its almost clear on the low setting image what are we looking at. the high well mobile phone camera quality

[Fraser]

EEEEEEK!  Adding noise = YUK !

[ataradov]

Adding noise before decimation (reduction of resolution) is a common technique.

[Fraser]

I only work with the full unadulterated FPA resolution 

My cameras are an artificial noise free zone ! (I hope)

 

[tesla500]

Adding noise before decimation (reduction of resolution) is a common technique.

What does this accomplish?

[M. András]

Adding noise before decimation (reduction of resolution) is a common technique.

What does this accomplish?

more sales in their higher software optioned models?

[ataradov]

Let's say you have signal with a maximum just a bit below your lowest resolution level. It will show as a plain color showing lowest possible level. Now add a bit of white noise. Pixels that were just below the threshold are now part time above the threshold, so you have two discrete levels. Pixels that were way below the threshold will stay there with higher probability.

Decimation filter will average the noise, so it won't show as much in the final result.

This technique is also used to "increase" resolution of ADCs.

PS: Proper name of this technique is dithering. 

[mikeselectricstuff]

I now have a much simpler hack method that does not involve service mode, and does not modify any existing files on the camera. PM me if you have a camera and want to try.

It is fairly likely this may work on the i3 as well - not sure how similar the i series is to the E
 

[tom66]

Also known as supersampling. I'd imagine that your typical FPA produces enough noise that adding more isn't a big deal.

[ixfd64]

I wonder if similar hacks can be applied to Fluke cameras.

[bean_dip]

I now have a much simpler hack method that does not involve service mode, and does not modify any existing files on the camera. PM me if you have a camera and want to try.

It is fairly likely this may work on the i3 as well - not sure how similar the i series is to the E

So much respect! I can't wait until my FFC cable gets here...I got the shipping confirmation from DigiKey today and they charged sales tax which means it should be shipping from inside the state.  Here's hoping it will be waiting for me tomorrow when I get home from work!

[mrflibble]

There were a couple projects trying to reverse-engineer bitstream format for Xilinx FPGAs. It was shut down by Xilinx lawyers almost immediately as it became public. It is still possible to find a PDF describing the effort, but referenced files are all gone from public view.

I take it debit from Jean-Baptiste Note on http://www.ulogic.org/ was one of those projects? Looks like his debit project has found a new home. From what I can find it looks to be the same codebase. Checking the source clones might also yield nice new resources.

git clone https://code.google.com/p/debit/

You know you want to...


edit: die stupid url tag, die!

[ataradov]

I take it debit from Jean-Baptiste Note on http://www.ulogic.org/ was one of those projects?

Yes, it was one of the projects. I found it when I was trying to do reverse-engineering myself. But seeing how lawyers are VERY active with cease and desist notices, I lost all interest. Completing and maintaining the project would be a lot of work, no reason to do it and know for sure that it will be shut down as soon as it gets any traction.

PS: Cloned it just in case

[Psi]

i reckon the noise seting is just to make the image quallity lower than the better models.

Would need to check the value on a E8 to confirm

[mrflibble]

I wouldn't be surprised if the noise setting is a bit of both. A lower (but non-zero) amount of noise might actually improve the image quality, as per the dithering remarks. And a higher amount of noise might be used to artificially make the sensor in lower end models appear crappier than it is.

[amyk]

I wouldn't be surprised if the noise setting is a bit of both. A lower (but non-zero) amount of noise might actually improve the image quality, as per the dithering remarks. And a higher amount of noise might be used to artificially make the sensor in lower end models appear crappier than it is.

Noticing how high Mike could turn up the noise from the E4's default, maybe they were thinking of making an E1 ~ E3 at some point.

Yes, it was one of the projects. I found it when I was trying to do reverse-engineering myself. But seeing how lawyers are VERY active with cease and desist notices, I lost all interest. Completing and maintaining the project would be a lot of work, no reason to do it and know for sure that it will be shut down as soon as it gets any traction.

The limited (public*) efforts to RE FPGAs is as you said, largely due to political/legal instead of technical issues; due to how the devices are constructed their bitstreams have a very regular format so it's rather trivial to figure out. Now if only someone does it anonymously and releases like Wikileaks...

*because no doubt the Chinese have done it out a long time ago, and have just not shared their results, at least on the English side of the web, maybe for these legal reasons.

[mikeselectricstuff]

I wouldn't be surprised if the noise setting is a bit of both. A lower (but non-zero) amount of noise might actually improve the image quality, as per the dithering remarks. And a higher amount of noise might be used to artificially make the sensor in lower end models appear crappier than it is.

Noticing how high Mike could turn up the noise from the E4's default, maybe they were thinking of making an E1 ~ E3 at some point.

No - it's just a variable that can be set to stupid values.
Having used it some more, I think it's probably mainly to reduce visible banding on smooth gradients - this may be more noticeable at lower resolutions. Might be interesting to see if the E8 has a different setting.
I think a small amount of noise does give an illusion of more crispness - the noise-free version looks smooth but subjectively slightly less well focussed. I wonder if a factor is making a cheap lens look subjectively better.

[Fraser]

Mike,

Here is the link to the cheaper Thermal Camera lenses that I mentioned previously. Maybe FLIR are using these in the lower end cameras ? Note that these lenss were first detailed in May 2009 and not the date shown on the articles. As such they could easily be in production now. Performance appears inferior to Germanium but in a low end TIC maybe that would not be such an issue.

http://www.gizmag.com/cheap-infrared-lenses-fraunhofer/23659/

http://www.en.iwm.fraunhofer.de/news/details/id/591/

As an insight into the crazy world of thermal camera optics, I just procured an 18mm, 45 Degree Germanium lens for my PM695. Its a large, multi-element, beast with very large lumps of Germanium in it of approx 60mm diamter. The price from FLIR and its agents is ...... over $7500    I have seen them cheaper on the used market though. The Germanium lenses still appear to be a very expensive component in the TIC.

Fraser

[Fraser]

An interesting document on the new technolgy for cheap thermal camera lenses is to be found here:

http://www.lehigh.edu/~inimif/eciworkshop/25D_Zhang.pdf

[olsenn]

When you say "cheap thermal camera lenses", do you mean the sensor as well, or just the lens? I'm assuming high-IR sensors aren't commonly found for cheap?

[Fraser]

Sadly not.

It would appear that FLIR have managed to reduce the manufacturing cost of the FPA (sensor), but we are uncertain as to whether the lens assembly has also been cost reduced in some way. I have presented a cheaper thermal camera lens that may have been used with the E4 FPA.
 

[olsenn]

It would appear that FLIR have managed to reduce the manufacturing cost of the FPA (sensor), but we are uncertain as to whether the lens assembly has also been cost reduced in some way. I have presented a cheaper thermal camera lens that may have been used with the E4 FPA.

Flir reduced the manufacturing cost compared to what? You mean it's confirmed that the E4 has a different lens/sensor than the E8?