Mµ Thermal Imager - real or fake?

[tom66]

Out of interest, what would be the cost, for example, for 160x120 vs 80x60? Could it be an optional extra without costing too much extra?

[Jay_Diddy_B]

Hi,

I would be interested to see an image of circuit board to determine the usefulness for electronics. The ability to focus at 4 to 6 inches is required.
Some thing like these images:

https://www.eevblog.com/forum/crowd-funded-projects/m-thermal-imager-real-or-fake/msg232013/#msg232013

Many thanks !!

Jay_Diddy_B

[free_electron]

"The firmware should be good to go"...  So that means they have images coming from the sensor right?  So why the hell have they not posted any?!? [Don't answer, I already know ]

The chef emerges from the kitchen to address the staving diners who have been waiting 3 hours for their meal, and speaketh...

"Don't worry folks, we have managed to get the texture of the glaze on the dessert fruit cup we're serving for dessert just perfect, you will really enjoy it!"

but the diners just want their damn fish and chips main course!

meanwhile the soup has boiled over and the potatoes are waiting to be pureed ...

[Hypernova]

Hi,

I would be interested to see an image of circuit board to determine the usefulness for electronics. The ability to focus at 4 to 6 inches is required.
Some thing like these images:

https://www.eevblog.com/forum/crowd-funded-projects/m-thermal-imager-real-or-fake/msg232013/#msg232013

Many thanks !!

Jay_Diddy_B

This, having used a Ti10 in my last job I'm ready to pony up $500 for 80x60 or $1k for 160*120 just for a well built sensor with say SPI output. On the one condition that it has close focus.

[Rasz]

The accuracy of the image geometry will suffer when compared to the Inframetrics lens

trivial to fix in post before displaying

[thermalguy]

Out of interest, what would be the cost, for example, for 160x120 vs 80x60? Could it be an optional extra without costing too much extra?

Since I have not completed the whole 320x240 to 80x60 system cost reduction effort, I cannot with any confidence give an answer. The bolometer however is almost a 4:1 difference in cost from a 160x120 to 80x60.

My intent is to start out simple and small. The problem here is that this is low volume at 2K annual. So details like flash, usb connectors and microcontrollers really aren't low price items as we do not receive any benefit from economies of scale. So a part that is $6 can have a cost loading of greater than 12% on the BoM. This isn't peanuts when your total PCB BoM price for the board needs to be less than $50.

Another aspect are peripherals. Just because a micro-controller has a built in USB peripheral doesn't mean it's free. Embedded USB libraries tend to consume at least 25% of the processor resources to implement. A micro-controller is a bad choice for this project in my opinion.

If USB is what is desired, then only the "USB daughter board" will have the USB components. If WiFi is desired, then only the "WiFi daughter board" with have the WiFi components. If we try to include everything in one shot, I think we are asking for trouble and a higher cost at these low volumes. If things take off then version 2.0 will be commissioned.

It is our goal to be able to sell each daughter card in the $40 range. That may be a stretch with the WiFi module but we will see.

We will make an Android App and will make it available to all along with the source code but to be honest we don't have much interest in specifically targeting smart phones in the beginning. We will leave that up to the user to find out what they want to do with it.

Our goal is to make hardware that works and is affordable. We will do an Android App and provide source code but that's about it.

[mikeselectricstuff]

Another aspect are peripherals. Just because a micro-controller has a built in USB peripheral doesn't mean it's free. Embedded USB libraries tend to consume at least 25% of the processor resources to implement. A micro-controller is a bad choice for this project in my opinion.

Microcontroller as opposed to what? CPLD/FPGA?
Not sure about others but some of the NXP Cortex ARM parts have a lot of the USB stuff in internal ROM. 

If USB is what is desired, then only the "USB daughter board" will have the USB components.

So how are you going to handle charging (assuming you're intending a standalone unit with battery) ?
A USB connector isn't going to cost a lot different to any other power connector. Or are you looking at Alkalines ?

[thermalguy]

Another aspect are peripherals. Just because a micro-controller has a built in USB peripheral doesn't mean it's free. Embedded USB libraries tend to consume at least 25% of the processor resources to implement. A micro-controller is a bad choice for this project in my opinion.

Microcontroller as opposed to what? CPLD/FPGA?
Not sure about others but some of the NXP Cortex ARM parts have a lot of the USB stuff in internal ROM. 

If USB is what is desired, then only the "USB daughter board" will have the USB components.

So how are you going to handle charging (assuming you're intending a standalone unit with battery) ?
A USB connector isn't going to cost a lot different to any other power connector. Or are you looking at Alkalines ?

Not sure what the USB "stuff" is that you are referencing but you mentioned ROM, which means on boot that code has to load into somewhere, so unless it is a co-processor you are going to have the same situation. Real time applications like imaging can be quite a task when synchronizing DMA channels and usually when using a high-bandwidth peripheral, it will eat up at least two channels. But I'm speaking from experience using 320x240 and 640x480 sensors, smaller array sizes will free up MIPS but only on the image processing side...core clock requirements.

We use Blackfin DSPs and Microsemi FPGAs here.

The charging question will have to wait until I have finished determining how much the BoM is reduced. It may be moved to the daughter card it may not.

[Hypernova]

What do you have in mind as the target audience? Since at this price point every cent matters how much is the core component going to cost and can you sell it with just that, the core (len+sensor+parallel data out)? I think for hobbyist level most of us just want the sensor and raw data (just like those Cypress based USB logic analyzers). Leave whatever processing you need to the PC and let the open source crowd work their magic. I also doubt throughput is an issue, the average hobbyists have been without access to thermal cameras for so long that most will probably be happy with even just 1 FPS. WiFi and what not is unnecessary fluff at this point, get the ball rolling first.

[mikeselectricstuff]

Not sure what the USB "stuff" is that you are referencing but you mentioned ROM, which means on boot that code has to load into somewhere

The USB stack. It doesn't have to load anywhere - it runs from ROM.
Remember 80x60x30fps is only 144kbytes/sec,and 4.8Kpixels, which is well within the capability of <<$5 microcontrollers to handle these days. If you need DSP there are chips like the dsPIC and ST Cortex parts with DSP instructions. Most MCUs would have enough PWM/timer peripherals to do stuff like generate timings etc. and as I said, USB would probably come for as close to free as doesnt matter

[Rasz]

We use Blackfin DSPs and Microsemi FPGAs here.

dont, all we want is a raw data dump from bolometer, everything beyond that can be done in software on tablet/phone/computer/Rasppi/any $50 arm computer/$20 mips WiFi router with USB host running openwrt

[mrflibble]

Remember 80x60x30fps is only 144kbytes/sec,and 4.8Kpixels, which is well within the capability of <<$5 microcontrollers to handle these days. If you need DSP there are chips like the dsPIC and ST Cortex parts with DSP instructions.

Pretty much this. For that kind of bandwidth you can easily use SPI to stream the raw data to your favorite MCU board. It all depends on how much standalone functionality you want to offer. Personally I would already be happy with the raw bitstream. I suspect most of the DIY crowd will want to do their own processing/connectivity anyways based on personal requirement. Personally I would connect it to an stm32 board + export the stream over tcp/ip. Plug it in to the router and good to go. Other people might want wifi, still others might prefer usb or bluetooth, who knows. Just spend the budget on getting good quality pixels out the door over SPI and you're all set IMO.

[mikeselectricstuff]

There will be stuff like fixed pattern offset/gain correction and shutter calibration, maybe some linearizing? - I suspect some of this may need cal data from a factory calibration process for each sensor, so it would make sense for the sensor system to integrate all this to provide clean image data.
Optional access to the raw data may be useful/interesting to some people,  but that's just a case of bypassing the processing.

Thermalguy - I think a lot of people would be interested to learn what steps are involved in processing the raw data from the bolometer to produce a thermal image, and how complex it is to drive the sensor - what sort of waveforms, voltages are involved etc.

[Hypernova]

There will be stuff like fixed pattern offset/gain correction and shutter calibration, maybe some linearizing? - I suspect some of this may need cal data from a factory calibration process for each sensor, so it would make sense for the sensor system to integrate all this to provide clean image data.
Optional access to the raw data may be useful/interesting to some people,  but that's just a case of bypassing the processing.

Thermalguy - I think a lot of people would be interested to learn what steps are involved in processing the raw data from the bolometer to produce a thermal image, and how complex it is to drive the sensor - what sort of waveforms, voltages are involved etc.

Cal data can be handled with a factory programmed SPI flash (or just on the on board MCU flash, really, how big can it get?) easily IMO, so some means of reading that from USB will be all it takes.

Strictly speaking you don't even need to agonize over the USB implementation, just get a MCU with fast UART, the TI 28335 I use at work tops out at 4.7MBaud and the chip has three of them. Pair that up with say the FT2232 dual ch USB to UART and you got at least 9.4MBaud of throughput at hand.

[mikeselectricstuff]

It is all pretty silly - if a Bad Guy in wants a TIC they can get one easily, but they are still classed as weapons due to outdated regulations, so it's more of a problem to legitimate users and manufacturers than it is to the people the regs are supposed to exclude. 
I skimmed through some of the regs and the only exemption I could see is that anything below 9fps is not covered. Didn't see any mention of resolution.
I think somewhere there is a threshold between "Illegal to ship to certain countries" and "Requires export license for any export" - I don't recall whether this is the same as the 9FPS threshold.
I've tried to get hold of Panasonic 8x8 sensors but Digikey won't ship them outside US (or inside US on a UK card) and UK distributors can't get them.

[Hypernova]

At least it's better than during the cold war when encryption routines were considered munitions. How the heck were they supposed to prevent people from mailing floppies out of the country?

[fcb]

Some encryption is still classified as munitions...

But what I found hilarious is that the £30 Digilent HS1 I purchased the other day from Farnell was classified as "dual use technology" and has export restrictions on it (still bloody useless to program XC95xxXL's!)

[AlfBaz]

speaking of restrictions, wasn't there or isn't there a restriction on US exports of MCU/CPU above a certain MIPS? Something to do with being able to do brute force decryption

[Rasz]

There will be stuff like fixed pattern offset/gain correction and shutter calibration, maybe some linearizing? - I suspect some of this may need cal data from a factory calibration process for each sensor, so it would make sense for the sensor system to integrate all this to provide clean image data.
Optional access to the raw data may be useful/interesting to some people,  but that's just a case of bypassing the processing.

Thermalguy - I think a lot of people would be interested to learn what steps are involved in processing the raw data from the bolometer to produce a thermal image, and how complex it is to drive the sensor - what sort of waveforms, voltages are involved etc.

Cal data can be handled with a factory programmed SPI flash (or just on the on board MCU flash, really, how big can it get?) easily IMO, so some means of reading that from USB will be all it takes.

Strictly speaking you don't even need to agonize over the USB implementation, just get a MCU with fast UART, the TI 28335 I use at work tops out at 4.7MBaud and the chip has three of them. Pair that up with say the FT2232 dual ch USB to UART and you got at least 9.4MBaud of throughput at hand.

This is what I was thinking, if FPGA/Blackfin DSP adds $40-100 dont bother, just expose raw data and cal info needed to process it. All it takes is few smart people to buy those boards and someone will write opensource stack to process from raw to nice false color images. Even $20 router has enough processing power to take care of that (400MHz mips cpu, 32MB ram).

[jancumps]

...
(Keep in mind that at FLIR headquarters, someone is very angry right now.)

My guess is that they don't give a damn  . They'll start to get angry when you release a commercial product with their embedded product hacked. And customers will start to get angry when their unit is defect and it can't be supported because the parts supplier doesn't support his hacked parts?

[EEVblog]

My guess is that they don't give a damn 

They shouldn't.
Anyone could have found out that info by taking them apart. And as Steve said, obvious when you look at the specs.

[Bored@Work]

My guess is that they don't give a damn 

They shouldn't.
Anyone could have found out that info by taking them apart. And as Steve said, obvious when you look at the specs.

But corporations work different. Often it is not about logic and facts. I have often seen managers making a mountain out of a molehill. More than once I had the impression they were doing it to underline their own importance and prove the reason for their existence (while they were just parasites) . And they focused on the particular subject, because it was the only one they thought they understood (but didn't).

[jancumps]

...
And I would take apart a $2000 thermal camera You think I can afford one of those?

Lease one

If car and truck manufacturers want to see how the competitors product looks like, they lease one at a local rental firm, do a teardown + rebuild, and bring it back  O0 .

[EEVblog]

But corporations work different. Often it is not about logic and facts. I have often seen managers making a mountain out of a molehill. More than once I had the impression they were doing it to underline their own importance and prove the reason for their existence (while they were just parasites) . And they focused on the particular subject, because it was the only one they thought they understood (but didn't).

I have cracking story about that...
1998:
I worked on the worlds best towed array seismic streamer. It's design and performance was all secret and we were all under NDA's not to reveal any info on the product etc.
It's a product that you can get a good idea of it's performance by simply looking at photos and things, especially the module that contained the hydrophones.
Well, the product was a big success and a huge export earner, and management entered it into the Australian Design Awards. They submitted photos, and all sorts of detailed tech info to the judges. We won, and the design and photos and info were published widely everywhere, and management were all chuffed.
http://www.gooddesignaustralia.com/awards/past/entry/slim-line-towed-array/?year=1998
(photos and story now missing it seems)

We still made contractors sign the NDA etc, and one day a new contractor phoned up management and went "WTF, I searched the internet and found this employee called Dave Jones who has all this technical information and detailed photos of the streamer plasted on his website. How can he do that and you tell me I'll be shot if tell anyone?"
When I came into work the next day there were a bunch of fuming management suits hauling my arse into a room.
I had to try and explain to them how I hadn't posted anything, I simply was proud that we had won the design award and linked to the public info on the design award website. It wasn't me who released this info, its was them!
Needed to say, this wasn't easy, as it was 1998, and not everyone understood how the internet worked.
They demanded that I remove all the information instantly.
Hmm, yeah, that took some explaining to try and convince them it wasn't my information, I didn't have access to the design award website, blah, blah.
But eventually understood I couldn't do that, but still ordered me to remove the link from my website, as if that would magically make the problem disappear. And it did because the contrator with his piss poor web skills couldn't find the website any more and reported as such.
They patted themselves on the back for the coverup of the info leak.
 

[mikeselectricstuff]

Here are some secret details about the FLIR i-series that nobody will EVER tell you without charging you $10,000 for a censored reverse costing report:

It runs Windows CE Core5.0. A notoriously easy to hack Windows distro (or so blackhat.com says).

Considering the moderate volume and limited functi9nality ot seems a bit nuts to use a 3rd-party OS in a product like that. I suppose with their near-monopoly they don't care about a few dollars o Microsoft on every unit.
 

Every single camera in the i-series has a 320x240 detector that is cropped, in software, to either 140x140, 100x100, or 80x80.

Assuming you're a good hacker, you could probably get access to all that 320x240.

It is possible that they use reject sensors with defects for the lower-res products, and/or possibly only calibrate them for the resolution they're sold as - I think the latter is fairly plausible if cal takes appreciable time as that equates directly to cost. thermalguy could probably shed some light on this.
   
..and probably 30fps, (though maybe the processing hardware may not be up to this) however depending on where you live you may run the risk of the  Spooks knocking on your door - as ITAR  regs treat TICs like arms, it's not inconceivable it could be treated like someone reactivating dummy firearms....

Displaying it on that sh!tty LCD screen would be a problem, and the camera also can't record video. But I bet someone could hack up a video output from some stray traces on the PCB.

It has an SD card slot....

This raises another question: how much do 320x240 detectors cost FLIR to make? Not much more than 320x240 CCDs cost, but they like jacking up prices.

The  vacuum encapsulation and germanium window would add quite a lot to the cost, and I believe the die size is substantially bigger than an equivalent optical sensor.

Who says I'm manufacturing anything? Here's the main component list, for anyone interested (mikeselectricstuff, I'm looking at you)

All donations welcome :-)

I have actually considered the idea of doing a kickstarter to buy something exotic for a teardown, but the ITAR stuff would make me slightly nervous of doing it with this.