Yet another cheap thermal imager incoming.. Seek Thermal

[bsliv]

Seek Thermal was on our local news today.  The Specialty Equipment Market Association (SEMA) car convention is in town.  One "expert" quickly reviewed about 10 products.  The second product, and his favorite, was the Seek.  No mention of the model, the price, or the competition.  He showed a pic of a hot car engine and a selfie. 

Why would Seek limit distribution of the SDK?  How much trouble would one get in if they posted the SDK?


WELCOME 2

ABOUT THE SEEKWARE SDK 3

THE SEEKWARE
MAKER API 3

THE SEEKWARE
MAKER LIBRARIES 3

THE SEEKWARE API 4

ENUMERATIONS AND USER DEFINED TYPES 5

ENUMERATIONS 5

USER DEFINED TYPES 7

SDK INFORMATION 8

FUNCTIONS 9

SEEKWARE_FIND 10

SEEKWARE_OPEN 11

SEEKWARE_CLOSE 12

SEEKWARE_GETSETTING 14

SEEKWARE_SETSETTING 15

SEEKWARE_GETIMAGE 16

THE SEEKWARE LIBRARY 17

DESKTOP PLATFORMS 17

WINDOWS 18

LINUX 18

OS X 18

MOBILE PLATFORMS 19

ANDROID 19

IOS 19

SAMPLE APPLICATIONS 20

WINDOWS SAMPLE 20

LINUX SAMPLES 22

OS X SAMPLE 28

ANDROID SAMPLE 28

IOS SAMPLE 30

SUPPORT 35

ATTRIBUTIONS 36

TABLE OF FIGURES 37

INDEX 38

[frenky]

If you share it with me I can guaranty you that I won't tell anyone. 

BTW:
Did anyone try to probe this signals coming from Seeks sensor?

[frenky]

I used marshallh image to show where sensor pins are accessible (pink-easy access, green - hard):
 

[frenky]

This info might be useful to somebody who's trying to understand Seek sensor.

JTAG:
I couldn't find where (if anywhere the other 3 pins come out):



EDIT:
Another 2 jtag pins can be accessible on the other side:

[frenky]

Getting very close to outputting clean images.

Most of the crazy jumping pixels are now under control.
Chart shows values of first 250 pixels for 1 hour. Then I pressed the "magic" calibrate button and recorded pixel values for another 15min.
The result is very promising so now I'm ready to start producing clean thermal images:

[miguelvp]

That's pretty awesome. Can't wait to see the resulting images.

[termal lee]

Franky, Seek should really thank you for this. we all should too 

[frenky]

Thank you guys for your support. I'll do my best not to disappoint you.

[frenky]

I tried outputting some images but I was not happy with the results so I changed a lot of code...

I decided that instead of trying to fix recorded pixels I'll try to predict future drifting...
So now I instead of waiting for an hour for sensor to stabilise I'm calculating drifting angle of each pixel so I can predict where it will be at next capture.

Results are now better (see below).
Left is correction of drifting with fixed value and right is with progressive value. I believe that this is a much better approach...
(Images were taken every 15 seconds)

[Fraser]

Impressive work Frenky

[arsenix]

Are those images of a calibrated thermal surface? (or something close ) I assume based on your previous post these are taken after the device has stabilized in temperature for a long period?

I'm not sure why the pixels would still be so unstable in that case. Maybe Seek's post processing is screwing it up. With an isothermal target image and thermally stable imager... the resulting image should really be pretty constant.

[frenky]

Yeah Seek was pointed to a thick aluminium plate painted with black matte silicone paint.
The drifting is just horrible... Even after an hour pixels are still drifting quite a bit.

[arsenix]

Could the temperature of the imager itself still be moving around with ambient changes? Just for the sake of science you could try placing it in a dark insulated box to protect it from ambient air and light.

About how much (percentage wise) are the pixels changing per frame? I really wonder if Seek's processing is adding to the noise in this artificial test scenario. Hard to say since they don't really let us in on what they are doing in the device.

[frenky]

I'm talking from my head but i would say that pixels drift for about 20 per frame. If pixel changes value from 8000 to 8020 that is only 0.25% change. But 20 is about 0.5*C so in "low contrast" scenes that is an issue.

Sent from my LG-D855 using Tapatalk

[arsenix]

How many bits are the values that get returned? 14 bits?

500mK of noise isn't bad if the accuracy is good as well. It isn't amazing either but I'd say anything under 1K noise is a whole lot better than what it offers out of the box...

[frenky]

Here is some info about raw values: https://github.com/lod/seek-thermal-documentation/wiki/Temperature-Values

Sent from my LG-D855 using Tapatalk

[encryptededdy]

I found with the Therm-App (which doesn't have a cal shutter) that when you change the temperature of the target object or the sensor significantly, the noise pattern / drift / nonuniformity changes significantly in amplitude. While the pattern stays the same, the amplitude changes pretty significantly.

[arsenix]

That is pretty typical. Each pixel in the microbolometer has a different offset/gain that is highly dependent to the device temperature. Full calibration is possible, but is not trivial.

This is a pretty research paper where they did just an offset calibration. The results show that the gain also needs calibration for good results:
  http://www.researchgate.net/publication/233864820_Method_of_detectors_offset_correction_in_thermovision_camera_with_uncooled_microbolometric_focal_plane_array

In this paper they came up with a model for the offset relative to temperature that held well to reality. Armed with this they were able to calibrate both offset and gain vs temperature without doing a full 2d matrix of device and target temperatures (which would take forever):
  http://www.coe.montana.edu/ee/jshaw/publications/Correcting%20FPA%20temp%20dependence%20-%20Nugent%20et%20al%20-%20OptEng%202013.pdf

[frenky]

I found with the Therm-App (which doesn't have a cal shutter) that when you change the temperature of the target object or the sensor significantly, the noise pattern / drift / nonuniformity changes significantly in amplitude. While the pattern stays the same, the amplitude changes pretty significantly.

Yes I noticed this. When Seek is pointed to isothermal plane the first frame after calibration is usually really clean.
But If I point it to a corner of the room which has 2*C less the noise really pops out...

[Alekov509]

Hi, everybody.

Watching this forum since last year end - Thank you for your deep investigation of the subject device. Here below some coins from my side for practical usage of Seek Thermal with Sony Xperia phone with side usb socket. Used for almost 10 month out of trouble.
Materials:
1. soft bumper case for the phone,
2. a piece of rubber core rope(abt 30mm lenght, 3 mm dia),
3. usb socket set micro for mobile . 
4. usb wire cord from any stuff - 2 inch lenght
5. a little piece of sheet metal(jerrycan etc)
6. 5min epoxy
7. 30-40 min of free time.

- Place the Seek at the most desirable(for you) position at the back of your phone,
- then pierce two holes in the bumper
- insert the strand and check the length, then cut,
- make two cheese shaped end washers from metal by scissors,
- fix ends of rope with washers,
- mix epoxy and apply to rope ends, cover with adhesive tape,
- insert the phone into the bumper, wait abt 1 hr,
- attach Seek,
- check the required lenght of usb cable by placing both usb male and female into respective socket,
- solder otg cable as required, then fix wires at both connections by drops of epoxy.

The benefit of short usb cable I've got with this - the cable may be stored inside the Seek Thermal suppled box.
Pls see pictures for deails.
P.S. another option discovered - I can hold the phone faced down untouched .

Hope above may be useful for any user having the phone with side usb connector.
Regards.

[encryptededdy]

snip

Off topic: You have one of those Sony lens style cameras...! (I assume). I thought nobody bought them.

[Alekov509]

That's right, QX10 has been gifted to me.

[in.Vitro]

Found this on youtube

[encryptededdy]

Still no emissivity adjustment :/

[frenky]

Update...

As said before pixel values stabilise after a half an hour or so. So when pointed to isothermal plane the images come out pretty nice and noiseless.

But when pointed to different temperature than the one of isothemal plane, each pixel gain is different. So the resulting images are very noisy.

This is the image of the room corner (aprox 2*C difference):


Taken with my app:


Taken with seeks android app:


Now that I understand Seek sensor much better I'm also able to get best possible image from Jadew's app:



As you can see. Jadew's app is still better because it deals with different gain of the pixels in a way that noise is not visible...

So where to go from here... I love thermal cameras but I'm done with trying to improve seek.

I'll rather spend my free time finishing a few other project which will bring me  some €€€ so that I can buy ThemalExpert from i3system when it comes out in December.
When I get it I'll put all my effort in developing best custom app for it...