Yet another cheap thermal imager incoming.. Seek Thermal

[casper.bang]

Ok so I just received my Seek after my wife came back from north america. Apart from having to hunt down the .apk because I'm in Denmark and not the US, it works like a charm.

As other people pointed out, the poor resolution, lackluster focus and high noise-level does make it strictly a consumer product. My unit also suffers from the gradient issue and as Mike mentioned is probably a QA issue. Even if that's somewhat annoying, at the price I think it's not too bad. It seems pretty clear though, that because of this, the user will have a hard time using it for any detailed work where the overall temperature difference is only a few degrees. This means using it outside your house looking for insulation issues takes some heavy interpretation on your part, as one single picture is not to be trusted (see House.jpg).

If I let the device rest, the gradient is minimal for the first 20-30 seconds but then its down hill from there. Whatever the problem, I get the feeling it's unlikely to be fixable by any software or firmware update, so unless you really want it now, it may be a good idea to wait for the second iteration when they ironed out the worst bugs.

Of course, should Mike or another smart guy find out how to mod it to fix the problem, I would be very tempted. Going to keep an eye on this thread

[efahrenholz]

@casper.bang,

At this point, I'm not sure we will see a version 2.0. The president of Seek passed on the 10th. I'm not sure how the company will continue, who will take command. My understanding is that it's a small startup. Not many employees. They may bounce back, but no definite direction or public announcement has been made yet.

Someone happened upon his obituary. It hasn't confirmed yet, but likely it is truth. There's no reason to fake a death of someone like this.

My best guess to things at this point is that nothing will change. Product will continue to ship out, people will continue to complain about the camera, and apple, but then decisions have to be made on if the company will close, if it will be sold, or if it will be run by someone else.

[efahrenholz]

Update: I have found information that Timothy Fitzgibbons may be the new president of Seek Thermal. Acker was the president and CEO, but I believe Fitzgibbons will take the position. This is speculation, but they were partners at Indigo and started Seek together.

[encryptededdy]

Seek seems to have a nice sensor on their hands however I would love to see it in a higher end device. I wouldn't mind paying the same price as the FLIR One for something like this (assuming gradient issues are fixed and noise can be reduced), and assuming the issues we're seeing now are due to lens, calibration or board design issues, not the sensor.

$350 for 200x150 is still a very good price

I think Seek has targeted too low of a price and may have made some compromises that make this camera unattractive for me.

[miguelvp]

With a proper lens and the same other things you'll be looking at a $500+ camera not $350.

[Uho]

To reduce  of the gradient must be divided into two the board. Processing circuit board and board of matrices with a temperature sensor. Put the lens of germanium. The lens is made of metal. It's expensive. I could not find a manufacturer of lenses for my project. A small number of very expensive. Need a high precision equipment. I was able to make only the third attempt. I do not think that the manufacturer wants to spend money to improve.

[mikeselectricstuff]

I think Seek has targeted too low of a price and may have made some compromises that make this camera unattractive for me.

Especially as they originally announced at $250...

[mikeselectricstuff]

At this point, I'm not sure we will see a version 2.0.

A lot of money has been invested in the sensor, so that's not going to be abandoned.
I do wonder how long it will be before the "meh, thermal imaging" attitude sets in - after the novelty wears off I'm not convinced there is a viable ongoing market for low cost/low performance consumer imagers.

I think Flir have maybe got it more right in going for an easily integratable module with all the magic inbuilt to provide a low entry cost to someone wanting to use thermal imaging in new applications, though their 1K MOQ is an unnecessary barrier to entry.

I wouldn't be surprised if Flir don't even bother doing Flir One for other phones, though it wouldn't be much effort for them to do a Seek style dongle for Android, so maybe worth doing that. 

[casper.bang]

At this point, I'm not sure we will see a version 2.0.

A lot of money has been invested in the sensor, so that's not going to be abandoned.
I do wonder how long it will be before the "meh, thermal imaging" attitude sets in - after the novelty wears off I'm not convinced there is a viable ongoing market for low cost/low performance consumer imagers.

I think Flir have maybe got it more right in going for an easily integratable module with all the magic inbuilt to provide a low entry cost to someone wanting to use thermal imaging in new applications, though their 1K MOQ is an unnecessary barrier to entry.

I wouldn't be surprised if Flir don't even bother doing Flir One for other phones, though it wouldn't be much effort for them to do a Seek style dongle for Android, so maybe worth doing that.

I don't follow the business side of things and IANAL, but it occurs to me that now such a relatively nice sensor has been developed, it won't just go away! Seek might get bought, IP may get transferred and the price may go up. However, I've seen enough of american capitalism, VP and startup forces in California to believe this will get dropped - after all, there's quite a distance up to the better Therm-App.

The Seek Thermal isn't perfect, but they got many things right I think - In Europe where energy cost is 5x that of the US, there is a big marked for a commodity device like this as it's within a christmas/b-day budget and most people have an Android phone laying around. For instance, I have identified 2 out of 3 air ducts in my house which is due to an upgrade (add a shutter membrane) because of the heat being lost (see image below). I wish it didn't have that gradient issue, but I am still impressed.

[mikeselectricstuff]

A few observations :
The diode on the back is definitely sensing temp - shunting it causes an instant change in temp readings, but no change in the image, or any change after a shutter close, so presumably being used to calculate a delta between the sensor pixel value and actual temp.

Cooling the lens mount produces a massive gradient, with the bottom of the display showing cold. The lens is inverting, so this corresponds to the shutter opening side.

Seems little doubt that the sensor is "seeing" the mount rather than only what is coming through the lens.
Again this could be lens alignment, but could also be that the lens is producing either too small an image for the sensor.
 
Sudden changes in the temp of the die or lens holder make it go very blocky, which I think suggests it is using the distributed "dark" pixels as a reference

Heating the back of the sensor seems to make little difference to gradient

 

[Fraser]

I was not under the impression that SEEK thermal bankrolled all the development of the new microbolometer so other companies may be able to buy it. FLIR are not a company to miss an opportunity so I do not see the new FPA disappearing any time soon. I just hope that SEEK Thermal continue with Roberts dream of TICs for the masses. If the image quality can be improved there is a massive CCTV market waiting for lower priced thermal security cameras.

[jaybeez]

Cooling the lens mount produces a massive gradient, with the bottom of the display showing cold. The lens is inverting, so this corresponds to the shutter opening side.

Seems little doubt that the sensor is "seeing" the mount rather than only what is coming through the lens.
Again this could be lens alignment, but could also be that the lens is producing either too small an image for the sensor.

Does this gradient you're producing through cooling match the position and scale if the inherent gradient?

Just going back through older photos from the preproduction days it appears something changed with seek's lens or bezel. the units or photos used by the early reviewers seem to have something else around the lens area then what is being made now. the large opening in the bezel for the tiny lens with the thin plastic ring taking up the extra space points to some kind of change somewhere along the timeline.

[Fraser]

Possibly a cost saving too far ?
I previously commented on the danger of making changes in the signal path when moving from pre prod to prod stages. A pity if performance has been compromised for the sake of 50 bucks rrp reduction.

[casper.bang]

What exactly is the purpose of the shutter? In normal photography, the shutter is used to control exposure time - regulate the amount of light for optimal picture (not too much, not too little). The shutter inside the Seek seems to fire every 2-3 second in an effort to do some kind of temperature calibration - presumably against a surface emitting as little infrared radiation as possible?! Why would this change so frequently over time though to fire so often?

[Fraser]

Google thermal camera "Flat Field Correction" and "Non Uniformity Correction".

The FFC shutter on other cameras operates approx every 2 minutes or longer. The SEEK appears to have a very unstable microbolometer and this causes the regular FFC shutter operations to compensate. The time that the shutter obscures the signal path does seem shorter than other cameras though so is not as noticeable to the user.

Aurora

[efahrenholz]

Calibration frames (NUC) are 4 frames. On higher end cameras, the NUC event is probably longer because the camera is stacking more frames to create an average smooth image. I also think Seek combines the NUC images. So after each NUC, the previous image is added to the new and the total taken is divided. I don't know how many they average in, but I think its two or three. If you leave the shutter open during a NUC, then allow it to work normally after, you will see the first image persist into several sets of calibration images.

[eneuro]

The time that the shutter obscures the signal path does seem shorter than other cameras though so is not as noticeable to the user.

Is it any noticeable sound when shutter is active in this Seek cam or this event is completelly transparent to user?
It looked like there were nothing to limit shutter position inside lens holder, so it have to hit somehow to stop moveing or maybe there is special controll signal to make this movement smooth?

Did anyone tried to catch on scope those shutter calibration pulses?

[mikeselectricstuff]

The time that the shutter obscures the signal path does seem shorter than other cameras though so is not as noticeable to the user.

Is it any noticeable sound when shutter is active in this Seek cam or this event is completelly transparent to user?
It looked like there were nothing to limit shutter position inside lens holder, so it have to hit somehow to stop moveing or maybe there is special controll signal to make this movement smooth?

Did anyone tried to catch on scope those shutter calibration pulses?

Yes there is a very audible click

[casper.bang]

Would it not be interesting to see what happens if you let the calibration happen by moving the shutter in FRONT of the lens rather than behind? With some luck, any imprecision of the lens (responsible for gradient?) would get adjusted/normalized by the calibration?!

[frenky]

Non-US forum members might be interested in this.
The Amazon has Seek thermal module on stock:
http://www.amazon.com/Seek-UW-AAA-Thermal-Connector-Black/dp/B00NYWAHHM

I just ordered one to my shipito.com US virtual address.
Total price on amazon was 240$. DHL Express to europe from shipito warehouse will be additional 30$.
If our customs don't interfere with the package I should get it for 216€ in total. 

Can't wait...

[eneuro]

Non-US forum members might be interested in this.
...
Can't wait...

It depends how many those units were manufactured already and if they plan to change/upgrade hardware a litle bit to make this sensor maybe more stable and resolve gradient issue in hardware, rather than in software.

Software solution probably I've already found, while just a while ago managed how to use OpenCV matrix operations to perform approximation of Seek Thermal calibration frames in realtime while this approximation is very fast (no FFT needed, etc) 

This is linear approximation by plane made in my software based on this attached sample Seek Thermal calibration single frame (this is 16bit PNG so imagej2 can read this image if someone wants to play with it-it looks like black only, but this is that is degraded to 256 colors grayscale palette when viewed in forums and other softwar e without support for 16bit depth grayscale channels).

Code: [Select]

$ file sts_cal-approx.raw.cal.000013.png
sts_cal-approx.raw.cal.000013.png: PNG image, 208 x 156, 16-bit grayscale, non-interlaced
After running software processing Seek Thermal calibration frames with image frames and joining them to create output image:

Code: [Select]

$ ./seek_thermal_opencv_test sts.raw.img.000011.png sts.raw.cal.000013.png
we get on its output approximation of provided calibration frame (this is from earlier router sample raw data provided in this thtread):

seek_thermal: Approximation of calibration frame: -0.015972115*x -0.015298731*y +1.693952814

Now, we can try to estimate tilt angle and try to correct this gradient issue while we can detect that approximated calibration surface is not flat (has gradients), so we might be able add some gradient correction to our image processing engine.

NOTE: Probably non linear approximation will be better depending on what causes this gradient to apear.
I bet this is sensor thermal instability, so will play with other nonlinear approximation surfaces too, but maybe decent output image improvement we'll have using this simple linear aproximation I've shown above, while it is very easy detect overall gradient based on this plane equation (normalized vector perpendicular to surface parallel to XY bottom plane should be close to [0,0,1] if we had perfect flat calibration frame with random noisy oscilations in pixel values) 

I will calculate later what this tilt angle can be in this case, but it is clear that there is small diagonal gradient while those coeficientf at x and y is not 0 ...

[blackboxdisease]

I plugged in, and used SEEK in a 10c room. Image quality was very very poor.
Placed my SEEK in the freezer for a couple minutes, then plugged it into my phone while still in the freezer. The gradient was apparent in the freezer from top right (cold) to bottom left (hot), holding the phone in it's normal position. (usb facing down). There was only two significant colours being displayed one of which was deep blue that I am going to say it is using as it's baseline. I let it grab a few frames, then took it out into my very cold room and image quality improved significantly. I then took it into a room of 25c and image quality was improved again. It was able to better distinquish the range of cold from hot. The lowest detected temp in the room was deep blue, and the hottest, being the baseboard heater, nearly white. I did not notice any linear gradient from top right to bottom left, until the thermal scene changed, but it only done so when the coldest detected temp was in the top right area of the screen.

I placed the camera in the warm room and connected the camera. The image quality was less in the warm room, and degraded more when moved in the cold room. I also noticed that there was an overlay of unchanged color (pixels) when moved to colder climate, making the image unrecognisable.

I subjected the area of the camera that was showing hot to a temperature much higher than it's initial detected temp at boot time and what was once blue in the top right, was now showing orange and the bottom left starting to show blue. This would logically be because the bottom left part of the sensor is effectively colder than the top right. So at least the colour part of the gradient has been reversed. I was also able to make the gradient disappear all together in the instance of extreme cold, and extreme hot. It is my assumption that the sensor is detecting radiation from the chipset and the gradient reflects the dissipation of radiation as it travels across the sensor as well as being diluted by the actual external detected radiation.

Since mike already has his disassembled, I would like to see him thermally isolate the chipset and pcb from the sensor using whatever method suits him. I was thinking of a temporary method by first masking off the pcb and chipset with tape and applying caulking, exposing only the temperature sensor. Or perhaps someone that doesn't have the lens assembly off, to just apply caulking around the part that is closest to the chips.

Of the 5 images, good_door.jpg was taken when the camera was placed in the freezer, bad_door was when the camera was let to sit in a room of ~25c and bad_gradient.jpg was when half the camera was subjected to ~80c at the side of the coldest part of the gradient, which I now know is flipped during processing, reverse_gradient is when the opposite side was subjected to very high heat and lastly, desired.jpg was when heat was equal across the entire sensor, higher than what it could detect from the chipset.

[miguelvp]

Didn't get to do much but I'm going to release the current version, I think it's stable enough. Note, I did upgrade my compiler to Visual Studio Community 2013 because it's free and I have the project under source control using the free 25 seat perforce, hopefully it doesn't cause much trouble compiling it with visual studio express.

link to post about VS Community 2013:
https://www.eevblog.com/forum/chat/visual-studio-community-2013-a-full-featured-ide-free/msg550822/#msg550822

Don't try to use the sliders on a win8 tablet, it's a nightmare to use, tried on mi wife's (well, now mine) dell venue 8 pro tablet.

As for bad images etc, maybe mine is a better camera, I went around the house taking pictures so it took me more than 2 minutes and it was working ok for my needs (checking cold spots around the house)

Also I took one of the radiator and it was spot on compared to a UNI-T UT301C IR thermometer, then one of the soldering Iron and it reads lower than the 350c I had it set at, then again it's a tenma so expected. The UNI-T couldn't get a good reading even if I had it set to 0.7 emissivity.

I'll post the images later,

Full compiled executable of attached source is at: (windows 7 and up)
www.miguelvp.com/ForMike/Seek

As is without trying to use an external reference (which it doesn't work as well as before using frame 1, sorry about that) 15000 is about 0 degrees celsius and 22208 is about 163 celsius. Not sure if it's linear didn't experiment enough.

Edit: another data point, 52C is 17226, so I get:

(22208 - 15000) is to 163
like:
(17226 - 15000) is to X

X = (2226 * 163) / 7208 = 50.338 kind of close

Then again the 0 Celsius where ice cubes coming out of the freezer, will measure it with the IR thermometer
I did and they are lower than -10C, but I don't know how long I had them out, but it looks linear enough with just 3 data points.

One more:
16561 is 36.4  C
With the above ratio I get: 35.3 C

Edit: and one more:
26500 is 260 C
With the above ratio I get: 260.06 C, so it does ok at high temps

Math: (( 26539 - 15000) *  163) / 7208 = 260.940 C, Ok shouldn't have round down, but still pretty close.

Attached picture of the 260C halogen overhead lamp that the UNI-T says it's 260C

[miguelvp]

Here is a picture from the Seek camera of my soldering Iron in it's cradle and the seeker output in automatic range.
The iron is supposed to be at 350C but since it's a Tenma not sure what it really is, the seeker image does better to show details in the iron because it doesn't over saturate.

Also I'm attaching an electrical oil filled radiator, that I measured with an IR thermometer and it gave me the same reading.

After that the kitchen door and all it's leaks. First one you can see my jacket hanging on a chair, not a ghost.

Then the pantry, bathroom & bedroom windows

And finally one of the living room windows that is double panel but it has an aluminium frame.
Edit: yup, that's me, and some plants on that one, not ghosts

I really can't fault it yet. I'll try in a bit to go outside and see if I get any good images from outside the house.

[marshallh]

Great job.

Looks like I have a clumb of pixels not fully detected. Had to use manual scaling