Yet another cheap thermal imager incoming.. Seek Thermal

[frenky]

To find drifting pixels you can use code like this:
Int[] Bellcurve = new int[51];

Loop trough all values...
Bellcurve[imgval[n]/1000]++;

The result is an array showing how much values is in each range:0-1000, 1000-2000,...

[jaybeez]

So I've been adding some more information to display the max and min values of both the sensor raw frame and the calibration raw frame, ignoring values under 2000 and over 49152

Seek Thermal sensor has 14bit output or more bits while @ 14bits maximum unsigned value is... 16383 ONLY so 50k value is 3 times bigger than 14 bit maximum, but Ok we'll see what I get from this raw data and compare

Anyway watched again this Seek Thermal gradient issue teardown by @Mike and made those screenshots for some investigation, while it is cleary visible that this shutter looks smaller than sensor external top dimensions and it looks like without lens holders its end position is not limited and one sensor edge formed some kind of triangle not covering this sensor uniformly (selected shutter edges and made it more visible by adding more green).


*snip*  maybe this affects temperature and that is why it is not uniform and we land with those strange gradients? 

*snip*

I think you are onto something. its does look like the shutter leaves a sizeable corner of the sensor uncovered when it hits the inside of the holder wall. that corner is also where my gradient starts (assuming non inversion).

What if mike's adjustment of the lens housing allowed the shutter to close differently?

his first mod removed a slight amount of material. after he tested the gradient was somewhat improved, but he attributed that to cleaning the lenses. what if that allowed the shutter to travel further into the housing?

his second mod slightly repositioned the housing and made even more improvement, to which he attributed the respositioning of the lens. what if moving the housing allowed the shutter to travel even further in?

my gradient appears to be a growing cold spot that increases with each calibration. maybe caused because the shutter isnt covering the sensor properly and it grows exponentially to a maximum or average? coukd reshaping the end of the shutter flag be the "fix"?

[miguelvp]

Well, other than finding those rogue pixels that interfere with my calibration I can get a pretty decent image after calibrating it with an external flat field and ignoring the internal one.

External calibration image (reference frame with camera facing down)


Internal calibration image (shutter)


I can also add sliders to change the image contrast and override the computed Max/Min so the rogue pixels won't make a difference at all. (other than show themselves in the image)

@frenky, I know about bell curve distribution and I was going to add a plot on the bottom of the image but I rather know the location since I do know already that it's happening.

Edit:
And it was a bug:
I changed:
                    v = v - c + 0x4000;

to:
                    v += 0x4000;
                    v -= c;

And it's happy and stable now, makes sense since I changed the original code from v being an int to being an UInt16

[Fraser]

@MiguelVP
Very nice work Sir.

I managed to buy an American version of the Motorola Moto G last week for only GBP26. I can now experiment with the SEEK hardware without risking my own Moto G.

I am just awaiting the SEEKs arrival from the USA.

[efahrenholz]

@miguelvp,

You publish your software anywhere? I wouldn't mind playing with that.

[miguelvp]

I will, I want to clean it up first and add the sliders to override the max/min

btw after 30 minutes with the same external reference image, it's still pretty stable (other than I bumped the camera a bit)

[eneuro]

I think you are onto something. its does look like the shutter leaves a sizeable corner of the sensor uncovered when it hits the inside of the holder wall. that corner is also where my gradient starts (assuming non inversion).

Updated this my post above and added two more detailed imaged from @Mike, but first video teardown BEFORE he cut this part of lens holder and later in issue gradient teardown it was not such clear visible 


reversed colors to show in Iron LUT colors black cold masked sensor part and getting hotter in time exposed sensor corner by applying more red  to show temperature differencies
Another side is protected by lens holder 

[efahrenholz]

Cool man, this is the first time I've seen anything this far along. This would make a lot of those windows 8 tablets a great hardware tool especially with contractors. Most are micro USB, and the camera clicks right in, touch calibrate to a piece of paper and pull your sliders to adjust contrast. Or auto contrast. Then a screenshot mode, email the shot to the client. Most tablets based on the newer Intel mobile chips are fairly powerful, could do some heavy processing on the fly. My Dell Venue 8 Pro would be a formidable tool with this on it.

[miguelvp]

As far as accuracy, I made a reference image using my home office wall, then pointed it to the upper right corner, the room walls are on the corner of the house (100+ year old house at that)

I used an IR thermometer to measure things:
Outside it's 9.0 C
Lamp (it's been off all day) it's at 20.6 C (room temperature)
Wall on the right is at 19.1 C where it's orange in the picture.
Corner is at 16.0 C
Total range of the scene is about 4.6 C.


So the sensor has the potential to be very accurate.

Also to note that the camera has been running non stop for one hour and I don't see any gradient forming after my external calibration.

@efahrenholz you mean like the one attached, not going to embed it since it's 800x1280

That's an earlier version of the code on the wife's Dell Venue 8 Pro (which she never uses so now it's mine )

[jaybeez]

Also to note that the camera has been running non stop for one hour and I don't see any gradient forming after my external calibration.

right right

edit to add : good. continue with this. it helps to narrow down the culprit. less likely an issue with lens alignment or heat radiating in through the slot, and more likely an issue with calibration, the shutter shape, or the temperature of the shutter.

[miguelvp]

Also to note that the camera has been running non stop for one hour and I don't see any gradient forming after my external calibration.

right right

Not really, that's actually the room wall being colder, here is a beer at 9 C (Kronenbourg 1664 because I really don't hate the French, although in theory it used to be Germany) anyways there is no gradient there. (same reference image as the one I took before and it's been running for an hour and a half already).



maybe you meant right right, as in "ok, alright" instead.

[ricksastro]

Superb work, miguelvp!    I'd love to have a copy of your code as well once you feel comfortable with sharing.   Awesome work you're doing!

[efahrenholz]

Also to note that the camera has been running non stop for one hour and I don't see any gradient forming after my external calibration.

right right

edit to add : good. continue with this. it helps to narrow down the culprit. less likely an issue with lens alignment or heat radiating in through the slot, and more likely an issue with calibration, the shutter shape, or the temperature of the shutter.

Or its what I suspect, which a thermal reflection. A slowly warming sensor is bouncing heat off the inside of the lens housing back at itself, which it can't account for if the shutter is blocking it during a calibration.

[miguelvp]

There is still some bug in the code, after 2 hours something made the upper value jump so it reads cold now


Note the Max value reads 19143 which is 2000 higher than it was before.

I was pointing at different things so some value made the Max on the image jump high and it got stuck there, the slider overriding will get rid of that problem, still not sure what happened.

I can always take another reference image and restore it. I tried that and it's back to normal without having to stop the program.

[Fraser]

MiguelVP

Looking Good 

Thank you for all the hard work you are putting into this. It is very interesting to hear that the microbolometer is more stable than it appeared when using the SEEK application.

SEEK Thermal, if you are reading this, please will you consider what MiguelVP and the rest of the EEVBlog 'development team' has achieved with your product and make suitable changes to your APP software. The SEEK microbolometer appears to offer better performance than your present APP suggests. That darned gradient issue is a real pity but MiguelVP appears to have found a decent solution that you could include in your APP....external FFC. If its good enough for the FLIR One, why not offer it on the SEEK  . I have used Inframetrics MILCAM professional cooled thermal cameras that use a lens cap and NUC routine for the identification of dead pixels. You could use a simple lens cap for the FFC. Cheap and easy 

I attach a picture of the MILCAM. The rubber booted button on the front panel adjacent to the lens activates the manual NUC routine and the user had first to fit the opaque lens cap to provide a flat field against which to check the FPA performance. It is a mechanically cryo-cooled FPA camera so this was just to identify dead pixels to the OS.

Aurora

[miguelvp]

Just want to point out that if I use the internal shutter I don't see the gradient forming either. It's just noisier and harder to clean up than using the external reference image.

But that can be cleaned up via software maybe using those first 6 frames, so it does point to a software bug to me.

[jaybeez]

Just want to point out that if I use the internal shutter I don't see the gradient forming either. It's just noisier and harder to clean up than using the external reference image.

But that can be cleaned up via software maybe using those first 6 frames, so it does point to a software bug to me.

that rules out a lot to me too, everything but software.

except that mike was able to eliminate it through physical alteration too.

[miguelvp]

Working on the sliders, for now I have them set from 14K to 20K and change them using the computed Min/Max, maybe I should just use them as deltas instead of absolutes

I think I need a larger range, this is an ice tray and a lighter:



I'm thinking adding an override button that toggles using the computed ones and using overridden ones starting with the computed ones when you press the button.

[efahrenholz]

@miguelvp,

I was curious, are you adding in the line 207 data to cancel out line noise?

[miguelvp]

Not yet, but I will eventually, so there is still room for improvement

Btw on the shutter gradient thing, If I just place the camera down using the internal shutter, it does have a ring around the center. Maybe the compensation software has a problem adjusting for it when the reference shutter gets hotter and the ring is more noticeable.

Since the ring is not centered it could be due to the lens placement and the software assumes the ring is supposed to be centered. Edit: so if the ring is due to the lens, then moving the lens to make the ring centered might fix the problem, but it should be able to be done by software as well.

I'll let it run for a while and take another snapshot in 20 minutes or so to see if there is any change from this one:

[miguelvp]

So after 15 minutes things did change, overall the image is the same, but the outside ring is hotter but so is the inside, but I still have some hot pixels that might be throwing things out of wack.




But if I point the camera to the router I don't see it affecting the image more than if I started that way.

[miguelvp]

With the modify range override you can see the lens centering problem better (if this is due to the lens that is)
 


I do need to change the LUT to be larger than 256 or scale the values of interest instead by using a linear interpolation.

Also I'm having some problems with modifying the range and sometimes it makes the app stop displaying, probably a divide by zero somewhere.

[eneuro]

A slowly warming sensor is bouncing heat off the inside of the lens housing back at itself, which it can't account for if the shutter is blocking it during a calibration.

This lens holder is metalic inside too, so shutter during calibration event removes part of the air inside and when leaves this sensor cage external heat is pumped inside from many air gaps in one corner side of the sensor, but whole internal lens holder will buildup temperature non uniformly.
Since there is additional direct radiation from the same corner side sensor during calibration event things can go even worse, while this thin shutter part which holds it sbigger end masking sensor will create additional shadow.


It is hopeless shutter design and so many ways to confuse sensor by non uniform IR from sensor cage inside and shutter itself not covering whole lens holder cross section above sensor, that for existing device only software update maybe will help hide those effects better and make them less visible on output images, but for never devices HARDWARE version should change to make this thing something better and more usefull than a toy 

Did anyone found any kind of temperature sensor on PCB in this Seek Thermal design?
Just looked to this thermally coupled ADT7031 12bit digital temperature sensor datasheet used in Flir 4 in such well designed PCB island and it has +/-0.5*C accuracy 0*C-50*C, +/-1*C -40*C - 85*C ...

NOTE: Images taken from @Mike teardowns.

[callipso]

Could someone (Mike?) take a HiRes thermal image of the bare sensor after it has heated up to the operating temp.? Maybe the die will be visible through the top window so we'll be able to see its size ond position and whether or not it is well covered by the shutter...

[eneuro]

@callipso  I could add also heat from one side corner in horizontal direction where die enclosure is exposed to direct heat  (below shutter working area above sensor) , while another part of this side slightly masked by lens holder as painted also in side view image from shutter perspective