Yet another cheap thermal imager incoming.. Seek Thermal

[efahrenholz]

The software isn't done yet, so the accuracy can't be called. They still have image quality issues to iron out, before they can even make any claims.

@bktemp,

In regards to the line noise reduction, you might try an argument that evaluates the pixel value first. Rather than a hard cut off line (where in hot zones it simply ignores the correction), try an expression that reduces the correction amount added the higher the pixel value is.

So basically take the min sensor value, subtract that from the current pixel. This gives us an effective range of 0 to %100. Then take that value and find what percent it is of max value. Lets call it pixelval. Then:

(pixelval*100)/(maxvalue-minvalue) = pixelval

Maxval-minval gives us the max in the range starting from zero. This equation gives us the percentage in the range we are at. Then take the value found in 207, divide it by 13 or 14 and multiply the difference of the value found above and 100. Basically:

Adjustmentval=((line207val/13) * (100-pixelval)) / 100

This gives you the number to add to the the original value.

So pixel(x,y)+adjustmentval = pixel(x,y).

This is going off a perfect linear scale. As the pixel value goes up, the correction value goes down. I don't know what the rate of change is as the pixel gets hotter, so the scaling might need a bit of tweaking... Not my best math, but I don't know how else to say it.

[eneuro]

There are 2 different scenes.   One is taken against flat notebook paper (flatpaper) at ambient 73F. One is an image of my router (router).

What about ambient temperature when you made those router images?
Can we assume it was the same and around 73*F ~ 23*C ?

It is interesting that when computed average of this 207th additional columns in calibration and image frame from those files you provided there with nice 16bit grayscale thermal router imaginery:
http://www.ricksastro.com/temp/seekraw/router_image1c.png
http://www.ricksastro.com/temp/seekraw/router_image2i.png
I've got average value 5078.125, then by strange conversion in maxima using http://en.wikipedia.org/wiki/Fahrenheit  conversion to Celcius I've got around 26, from 79, which is close to ambient 73*F, and is it possible inside Seek Thermal camera could be 26*C ~ 79*F ? 

Code: [Select]

(%i56) max: 2^14;
(%o56)                               16384
(%i54) tf: 5078.125/max*256;
(%o54)                           79.345703125
(%i55) cel(79.345703125);
(%o55)                         26.30316840277778
This means +3*C ~ +6*F difference between room ambient temperature and maybe those sensor zone thermistors formed in hexagon pattern "black pixels" ? 


It is interesting that those temperatures >23*C computed from this additional 207th column for each sensor raw image data (which are similar for calibration and image frame, so taken its average for each row) are above estimated room ambient temperature 22.7*C ~ 73*C , which could make sense, while we got temperatures from sensor greater than ambient temperature

BTW: Can someone verify this gueses based on a few basic measurements as described a few posts above?
Even one set of data with measured object temperature at given room ambient temperature (thermal camera should be there for longer time) and those calibration and image frame, could help solve this interesting temperature puzzle...

[bktemp]

@efahrenholz
That is basically the same as I did yesterday, except I did use the pixel values without the min/max scaling.
Your idea works great for some frames, but others fail completely. I think the scaling makes it worse without manual adjustments for each frame: If one frame has a hot or cold pixel somewhere, it changes the min/max value and therefore changes the gain applied to the 207th value. If you look at the animated gif I have made yesterday, you can see a few pixels that change wildly and are sometimes even the hottest/coldest pixel:
https://www.eevblog.com/forum/testgear/yet-another-cheap-thermal-imager-incoming/msg544366/#msg544366
Therefore I would go in the other direction and use the absolute values. Then the noise does not effect the horizontal line compensation.

The attached graph shows the raw value to temperature (in °C) convertion for a different thermal imaging camera. Lets assume the Seek has a similar curve: As you can see the sensitivity increases with temperature. This would explain why the compensation creates dark lines in hot areas: The hot area has an increased value compared to the lower sensitivity of the compensation applied.

[efahrenholz]

@bktemp, the min/max value should be constant. The frame max and min values shouldn't be considered as they can change the calculated gain on individual pixels per frame. We need to find the lowest and highest reported values the sensor can display. I hope that clarifies my math.

[KK]

Has anyone received the iPhone version of the camera yet? The app is out so that is no longer an issue.

[bktemp]

@bktemp, the min/max value should be constant. The frame max and min values shouldn't be considered as they can change the calculated gain on individual pixels per frame. We need to find the lowest and highest reported values the sensor can display. I hope that clarifies my math.

Ok, now I've got it. But that should not change much. Using the maximum and minimum values that can appear in the data you basically get the same as I have already used: pixel=pixel + (pixel+offset)/gain * 207th value
Instead of alculating the gain from min/max I have used offset and gain values. I have tried different values and subtraction pixel from offset instead of adding it. It was better than without the gain compensation, but not good enough.

Now I have tried using pixel=sqrt(pixel) as a coarse approach for linearization. The results look a bit better, but still far away from beeing perfect. The intensity of the remaining horizontal stripes change after each calibration frame. Maybe the linearization is needed before subtracting the reference frame.
By subtracting the shutter frame you lose the absolute temperature information and only get the difference to shutter temperature. Therefore you need to add the shutter temperature again (maybe using avarage value of calibration frame?) Or is some other temperature information available?

[mikeselectricstuff]

Has anyone received the iPhone version of the camera yet? The app is out so that is no longer an issue.

There claim that they are still having issues with Apple hardware approval

[Monkeh]

Has anyone received the iPhone version of the camera yet? The app is out so that is no longer an issue.

There claim that they are still having issues with Apple hardware approval

Flir bribe bigger.

[efahrenholz]

@bktemp,

Bummer well it was worth a go. It seems reverse engineering this puppy is not so simple.

When you say shutter frame, do you mean that you have the option to actually ignore flat field events? When I tried sgstairs source the shutter still clicked away, so I assumed it was impossible to stop this even from happening. Like its handled by the firmware. Hmmm I wonder if anyone has messed with that chip yet...mike?

[callipso]

Has anyone received the iPhone version of the camera yet? The app is out so that is no longer an issue.

There claim that they are still having issues with Apple hardware approval

Flir bribe bigger.

 

[eneuro]

The intensity of the remaining horizontal stripes change after each calibration frame.

How much does it change -more than 10% per minute? Did you tried to estimate this?

However, after watching this exellent @mikeselectricstuff Flir E4 teardown http://youtu.be/NtqUE67BUDI   and reading those related posts to this queensu: enhancement: extract binary data from FLIR radiometric jpg where people did amazing approach applying temperature info using their own scripts and utilities, I hope Seek Thermal makes all this complicated thermal calculations on its onboard PCB bulky MPU and maybe sends via USB preprocessed thermal information to minimize proccessing effort on customers devices, while they want this thing running on diffrent platforms.

So, I will deal in a very different way with those Seek calibration and image frames and i'd like implement something like this below for the moment and apply 16bit Iron LUT on limited output temperature ranges.

Inspiration for this attempt is mechanical construction of Flir E4 shown by @mikeselectricstuff in his teardown video linked above.
Starting from thermal scene with objects heat source we have in Flir E4 in a IR waves path  lenses via shutter hole and sensor shown below.
Seek Thermal shutter thermal design looks very crappy in comparision to this built in Flir E4 
Everythong mounted in solid metal case with temperarture sensor sensing its temperature. When @mikeselectricstuff frozen this temperature sensor on PCB Flir's output ended with completelly blanked display while this temp has gone below its limits   
Even sensor looks like is thermally connected with this solid metal (not plastic) case and shutter nicelly hidden in its case, while in Seek design it is simply floating in air inside PCB cover-I do not like it at all 

BTW: In my recent designs I've choosen now this MLX90620 Series Wide (60x15) 16 x 4 Array 3 V 9 mA Thermal Array Sensor - TO-39 while it has everything I need and is far below $100 for higher volume.

16x4 pixel thermal imager
Its +/-1*C accuracy in temp range 0*C-50*C and easy I2C interface make this device much more interesting for me in a few projects than Seek Thermal dongle without any user manuals and only few lines of text as datasheet 
I use OpenCV for MLX90620 post processing, so simply it is hobby attempt to see how Seek image processing would look like based on sensor USB raw 16bit data people there were able to capture-for the moment I do not implement any Linux low level drivers like for example v4l2 used with classic visual light cameras.
Seek Thermal investment should be to create such Linux drivers if they want to compete on thermal imaging market since there are many small PCb computers running ARM Linux like eg. BeagleBone I'm involved now in my projects.

[bktemp]

When you say shutter frame, do you mean that you have the option to actually ignore flat field events? When I tried sgstairs source the shutter still clicked away, so I assumed it was impossible to stop this even from happening. Like its handled by the firmware. Hmmm I wonder if anyone has messed with that chip yet...mike?

Since I do not own a Seek camera, my only source of Seek raw data is the captured USB traffic from marshallh. It contains 75 image frames, 7 shutter frames and 6 preshutter frames.
I have no idea what kind of preprocessing is already done inside the Seek camera. Since the Seek app only gets the same data, all information including the absolute temperature must be in there somewhere.

The intensity of the remaining horizontal stripes change after each calibration frame.

How much does it change -more than 10% per minute? Did you tried to estimate this?

Sine the only Seek raw data I have to work with contain only 87 frames, I can't give an absolute number. I did noticed some changes after the shutter frames. Not big, but noticeable. Maybe I am wrong and it is only the changing noise pattern that adds a different pattern to the stripes.

[miguelvp]

@bktemp,

Bummer well it was worth a go. It seems reverse engineering this puppy is not so simple.

When you say shutter frame, do you mean that you have the option to actually ignore flat field events? When I tried sgstairs source the shutter still clicked away, so I assumed it was impossible to stop this even from happening. Like its handled by the firmware. Hmmm I wonder if anyone has messed with that chip yet...mike?

There is a bit that tells you if it's a calibration frame or flat field event or not, you can ignore them but they are still going to happen since that is in control of the firmware.

Maybe there is some USB command that can be used to prevent it but as of now you can't prevent them from happening.

[efahrenholz]

@bktemp,

Bummer well it was worth a go. It seems reverse engineering this puppy is not so simple.

When you say shutter frame, do you mean that you have the option to actually ignore flat field events? When I tried sgstairs source the shutter still clicked away, so I assumed it was impossible to stop this even from happening. Like its handled by the firmware. Hmmm I wonder if anyone has messed with that chip yet...mike?

There is a bit that tells you if it's a calibration frame or flat field event or not, you can ignore them but they are still going to happen since that is in control of the firmware.

Maybe there is some USB command that can be used to prevent it but as of now you can't prevent them from happening.

I see, so it isn't automatically subtracting flatfield images out of the frames. That's up to post processing. Its just sending the calibration frame up and indicating it in the first dark pixel what type of frame it is.

[miguelvp]

That's correct, it's up to you to use it, throw it away or compare it to previous flat fields

Edit: by comparing to previous flat fields I mean you can make a program that shows if the flat field calibration frame shows the gradient drift.

[efahrenholz]

Interesting thing...as I thin I might know the source of my particular imagers gradient. Its the shutter, but its a hot shutter. My gradient comes from the side the shutter comes from, so I took a magnet and slowly moved it towards the direction to pull the shutter in. The shutter enters the field of view in the same corner the gradient appears from. I measured the temperature of the shutter when forced shut at around 108 after the seek has been running awhile.

The sensor is getting side radiation leaking from the shutter. A heat source is too close to the sensor and its showing. How the heck is the shutter getting so hot? Is it actually resting against the sensor?

[miguelvp]

The gradient is formed by a cold spot from where the shutter enters the chamber.

The chamber gets hotter over time and that might change the temperature of the shutter, but the cold coming from where the shutter enters seems to be the reason we are seeing that gradient.

I don't think it's the flag itself and I haven't looked if the gradient exists using the raw images yet since I haven't tried eliminating the banding just yet.

[Fraser]

I have written to SEEK asking for their assistance with my development of their product for other uses. I will be working with a standard bare PCB without the SEEK lens as I will be adding some specialist optics. The camera will be temperature stabilised and adapted to work remotely from the host. Sadly I have yet to receive a reply from SEEK Thermal so they may be too busy or unwilling to assist me in this development work. I will have to wait and see if they respond over the next few days. Plan B will be to order some more SEEK cameras when they become available in the UK.

Aurora

[efahrenholz]

My gradient is the opposite. Its not cold, it's hot. The shutter starts off about the same temperature as ambient. The longer it runs it gets hotter, and it seems to run about as hot as the sensor. People report the sensor at around 38°C and that about 100°F, which is close to what my shutter reads when forced shut.

[miguelvp]

My gradient is the opposite. Its not cold, it's hot. The shutter starts off about the same temperature as ambient. The longer it runs it gets hotter, and it seems to run about as hot as the sensor. People report the sensor at around 38°C and that about 100°F, which is close to what my shutter reads when forced shut.

That makes sense if you are forcing it shut since that's the temperature of the sensor.

But on normal operation some coldness comes from where the shutter enters the optical chamber. Even so, I think it could be compensated by software, either via the firmware or the app itself.

And Seek responded and they think it might be possible to address the problem via software.

[slackaz]

I got my camera a few days ago. I agree the gradient issues basically ruins the device. The only thing that seems to workaround is going to split-screen. Still not nearly a perfect solution. 

I attached some samples pointing at a wood desk.

Also, here a link to a video where I rotate the camera around my foot. The gradient is clearly visible and rotates with the camera.

https://drive.google.com/file/d/0B0byoSUUY33oM3JXYkNTNjM5Skk/view?usp=sharing

[mikeselectricstuff]

I have written to SEEK asking for their assistance with my development of their product for other uses. I will be working with a standard bare PCB without the SEEK lens as I will be adding some specialist optics. The camera will be temperature stabilised and adapted to work remotely from the host. Sadly I have yet to receive a reply from SEEK Thermal so they may be too busy or unwilling to assist me in this development work. I will have to wait and see if they respond over the next few days. Plan B will be to order some more SEEK cameras when they become available in the UK.

Aurora

I think they are way too busy with production to be spending time talking to people about special applications.
I'd be surprised if they didn't produce some sort of OEM core at some point in the future, after the beta testers  consumer customers have helped refine it to the level OEM customers would expect

[eneuro]

And Seek responded and they think it might be possible to address the problem via software.

They had to write something to keep people buying this since it is still in top google search "Seek Thermal gradient" issue 

It looks like that they haven't got even basic things in their software like temperature scale with thermal LUT overlayed 
Did they added this to its latest Andoid update and it is available now or still on things to do, but this should be done as the first UI feature
while it is difficult to find thermal software without such basic thing 

Even small thermophile arrays with a little postprocessing have temperature scale and as we can see below no problem to find person looking into this scale while colors match very nice environment 

But this tiny device has close to 50 pages datasheet with equations needed to calculated things with examples 

For the moment frustrated customers will listen to what will be done in... not defined future.
Simply it is not fair.
BTW: They should run their own forum at cover complete lack of any published documentation available for end users...

[miguelvp]

For the moment frustrated customers will listen to what will be done in... not defined future.
Simply it is not fair.

I did read the customer reviews and the main concern usually is that their device is not supported or they did install cyanogenmod which breaks OTG so it won't work.
I think only one customer mentioned the software needs more features, and I didn't see a single complain about the gradient (other than here or by people here).

Only frustrated customers I know off are the IPhone ones, since they don't have their device yet.

[Fraser]

@Mike,

I did receive a very friendly reply from SEEK last time I wrote so you never know, they may be more friendly than we think 

I am interested to see whether the performance of the camera improves significantly when a decent (larger) lens is placed in front of the new 12um microbolometer. I will also be lowering the temperature of the whole microbolometer assembly to see how the noise responds. Much will depend upon whether the SEEK firmware can cope with the hardware being enhanced. The SEEK is still the cheapest source of a decent resolution microbolometer and ROIC at this point in time. My application is non radiometric so temperature contrast is far more important than temperature accuracy.

For fun, I attach pictures of one of my specialist lenses that I will be experimenting with.....and yes it is a thermal mirror reflex type telescope of sizeable dimensions and magnification  Long range thermal imaging anyone 

Now if only SEEK will provide the hardware I need........


In the mean time I have a beautiful Bosch Metal Mickey dual tech thermal camera to play with ..... she's a lovely piece of engineering    I attach a picture of 'Mickey'. SEEK Vs Metal Mickey......Hmmmmmmm I think Metal Mickey wins somehow ! But then $20,000 buys a lot of camera.

(If you wonder why the camera is called Metal Mickey...Google Bosch MIC CCTV camera)