Yet another cheap thermal imager incoming.. Seek Thermal

[Uho]

You need to close the lens. And to reduce the temperature difference. This lower readings. Then you can see the gradient. You have a big difference. So nothing can be seen.

[efahrenholz]

@lunchbox,

You have a very clean sensor from the dynamic range in your photos. I tried using the windows app and my cameras noise level and dead pixels make it look terrible. The only things that look good are large temperature variations. Any scene where the temperature is less than 20 degrees of variation looks very noisy.

I know this is probably a weird direction to go in, but perhaps the noise level we are seeing on most sensors is from a bad vacuum in the die. I'll need a macro lens but I noticed a shading to the surface of the imager die. It starts in a corner, it might be an edge crack.

[Lurchbox]

1st pic is camera flat on brick wall outside
2nd is the subway during rush hour
3rd is a pic of city street with what looks to be a steam line underground
4th is of my shoes after walking in from the cold

[Lurchbox]

Combining layers in gimp looks pretty  decent. with and without edge filtering

[miguelvp]

Picture my flat surface do Seek.

One thing to take into account is that my version of seeker will spread the LUT automatically from min to max
16041 - 16256 doesn't seem like a lot of heat difference, but I never had the time to map the raw value with the actual temperature in Celsius or Fahrenheit. Maybe doing screen shots of pointing to icecubes will give us a reference what is -6C (if that is what an icecube is at when it comes out of the freezer)

Maybe measure it with a DMM thermometer or an IR thermometer so we can start working on a table to convert those raw values to an actual temperature.

But I'm swamped with real life work so I can't do those measurements or code modifications.

Edit: also as it has been stated many times, pointing down to a surface at a very small height the reflection will alter the reading so it's not a real good test. Pointing to a uniform floor at one meter height will probably give you a more true gradient even if the sensor can see itself on the reflection at that distance will make the reflection smaller.
 

[Uho]

I have done tests with a gradient at different distances. Including flat wall at a distance of 1.5m. The results are very similar. The actual temperature needed in the least. After the elimination of noise and gradient. Practical solutions no one was able to achieve. I saw a strange behavior Seek. Without shutter image of the object disappears after only a few cycles of the shutter. When the camera redirect to the object on a flat wall. I can not understand.

[frenky]

Combining layers in gimp looks pretty  decent. with and without edge filtering

I've been playing with image overlay too:

[miguelvp]

I really want to get back to this, specially now that I have a Jetson TK1 with Linux and 192 Cuda cores. But work is taking all of my time, and because it's winter a friend of mine asked me to borrow my Seek so I can't do much for the next couple of weeks.

BTW, about overlays, search for "image registration" that's the terminology used for taking two images from different angles and correlate them in image processing.

http://en.wikipedia.org/wiki/Image_registration

[Lurchbox]

Actually was just reading a similar PDF at: http://chsopensource.org/panoramic%20multispectral%20imaging.pdf
this one may also have some useful info.

pic of an electrical transformer and random city shot,

[Lurchbox]

Combining layers in gimp looks pretty  decent. with and without edge filtering

I've been playing with image overlay too:

Looking good 

[frenky]

I've been working on pan/tilt motorised head lately:

First version (didn't work well because motor axes would get loose.)


Second version (added alu plates with screws for fixing axes)




And the result:
(Left image is only 1 frame, right image is 20 frames stacked)


Edit added overlay on this new image:

[Seeker]

Another statistic: a consumer review from Amazon, indicates a gradient and temp error ranging from +3^oC one end to +13^oC the other...not too good. 
------
"I was very excited about getting this little camera. I had read about, and watched the YouTube videos about the gradient problems, but since not everybody complained about them, I took a chance, and lost! The left, or top 1/3 of the screen (depending on which way you turn it) always showed colder. I took it out to my big cathedral ceiling that is insulated with spray foam, and very consistent in temp. I first checked it with my inferred gun, and it was 60 degrees +/- 1 (I am retired and keep the house cool!). The camera showed much colder on the left 1/3 by about 10 degrees, but was still reading about 3 degrees to high. The sad part is, the major 2/3s of the screen was reading sometimes more than 13 degrees too high! I bought this solely to find any weak points in my insulation, but its was sadly totally useless to me, and is going back. Sounds like I am lucky I got it from Amazon, because Seeks customer service doesn't sound so good! I also had some small issues like the temp readings half off the screen, and having to reboot my phone a lot to get a pic to show again on the cam. Hopefully they will get the bugs worked out over time, because it could be really cool, and useful camera! On the plus side, the packaging was really cool, and so was the protective case. I will attach a pic of my ceiling that was all 60 degrees, and one of the lens, which doesn't look like American quality, but from watching the YouTube videos, probably not the problem."
--------

[Spyke]

Seek gets some competition... a useful Flir One for android, with an internal automatic shutter like on the other Lepton based standalone cameras.
Apparently they doubled the resolution of the Lepton as well? Can anyone clarify?

http://gizmodo.com/flirs-new-tiny-predator-vision-thermal-camera-now-fits-1677792125

[frenky]

It has been discussed here:
https://www.eevblog.com/forum/testgear/new-flir-products/

[frenky]

After a few hours of testing and tuning I think I have found the sweet spot for best images:
Left is original from seek app and right is 16 images stacked in gimp.


The last thing to do is to write code that will:
- offer quick preview (wihout steppers moving)
- take 16 images while moving steppers (4x4 matrix with each image 1.2 pixel apart)
- stack all the images and calc new one with median average and higher resolution 624x468
- display new image and enable saving it

[efahrenholz]

@frenky,

That is looking good. You really could use a 3D printer to rapid prototype some of the smaller pieces.

On a side note, that new flir lepton core looks like it might kill the Seek. I hate to say it, even with it still not at a higher resolution, the image quality is impressive. Seeks software engineers are slacking, I don't know if they are being tasked with customer service issues or what but we haven't had any news on the SDK, or even that software update that broke the temperature readings.  The Apple version was very late to the game, customer service was...meh. I personally want Seek to win. Flir needs a good ass whooping, but it looks like they might win yet again. If we could just get some good software for this camera...I honestly don't even know what direction anything is going. If they open source it, (which is probably never going to happen), the generous developer community would have most of these issues fixed. The sales would skyrocket. Everyone could use a thermal camera for...something. Hell, just imaging your urine stream is enough to sell it to someone. At this point, Seek will have to lower their price to make it appealing, even with its current fatal design flaws.

[Fraser]

I think my 'as new' SEEK camera will be finding its way onto ebay to recover some of its value. Nothing has been heard from SEEK Thermal for ages. I am losing confidence in them and their product. Can't wait to get my paws on a new FLIR One Android instead.

Such a pity.

Aurora

[miguelvp]

That's a pity but we even haven't tried to use the extra commands that are documented in the decompiled code including shutter events etc, I just don't have the bandwidth at the moment (nor the camera because I lent it to friends to winterize their house) but soon I'll have time to get back on it.

Looking at the pictures using seeker for windows gives me confidence, as far as Seek releasing their dev sdk it hasn't  been that long and it might be a mute point. Also we don't know Seek's margin on their devices.

One thing is for sure, there is a market for a micro usb reversal connector, a tiny double sided board with 4 vias or just short insulated wires power and ground on the back twisted on the front and conformed or potted will do.

Also Image Registration is not too difficult, I've done near infrared registered with visual sat images and register them with elevation data and further with street maps, for mapping applications using my own software but there are tools like matlab that can do it for you as well. It's just a matter of detecting the field of view, rotation scaling and you can even do the distance thanks to the stereo information, but we are a long way from that at the moment.

http://www.mathworks.com/discovery/image-registration.html

Edit: for example, this 3d render is produced from only GIS and DEM information. Not real time but the task at hand is simpler than this

[frenky]

@efahrenholz:
I agree that 3D printer would come handy. I actually have Prusa V2 90% finished (and collecting dust in an attic)...
Even better would be to have a 5 axis cnc.

About Seek...
They haven't posted anything since 23th of December. Perhaps PR person gave resignation...
And also no new development of seek app is not a good sign. I don't think we'll ever see a SDK.

Every major project (like Seek module) has to have at least one person 110% committed and passionate about it to be successful.
I'm afraid that that person was the deceased CEO od Seek. If this was the case then Seek is history...

I love playing with Seek camera while trying to improve it but I would swap it in a heartbeat for new Flir One android module with MSX and 160x120 resolution.

[Stoffel Janssens]

Hi All

I received the Seek Thermal and I'm pretty impressed.
Thanks to miguelvp I can run it on Windows!

Gradient
When I plug the device in, I see that the gradient gradually appears. Internal heating?

Cheers

[miguelvp]

Thank you but the credit goes mostly to sgstairs he made the C# Visual Studio framework with all the camera initialization, I only added the lookup table and processing and use of the 16bit data instead of the 8 bit data and scaling etc but without that framework and the usb protocol handshake I would not have been able to just jump into it so quick.
 

[TinWhiskers]

With previewing of the new Flir One at CES it may be more difficult to assess whether waiting for improved Seek software or for the Flir to come out makes more sense....for me, the limited temp range of the Flir One ( current model ) is a deal breaker ... Will be interesting to see if the new dongle improves on this issue .

[z01z]

A new version of the android app has been released (1.7.4)!
For me the gradient is gone .

[arius]

Wow, they really fixed it.

[Fraser]

Thanks for letting us know. I had given up checking for an update !

Time to download it and see what's what with the new version.

Aurora