Yet another cheap thermal imager incoming.. Seek Thermal

[Uho]

I waited three weeks from the date of payment.

[mitulpatel83@gmail.com]

Just received my Seek  For Android Saturday and was wondering if anyone has the old version of the app?

If you don't mind answering, when did you order, and how long did it take to arrive?

I ordered mine directly from Seek on November 18, and got email on the 19th saying it had been shipped. But FedEx tracking STILL says "Label created", meaning Seek hasn't turned over the package to them yet. I've sent three messages to orders@thermal.com, and I haven't yet seen a single reply (other than "got your email" auto-responses).

I guess I'm just looking for reassurance that they actually ARE shipping currently. From all I can tell, it looks like they're taking money, but not delivering products, and not answering email. I don't want to dispute the charge via my CC, but it seems like that's the next step. 

Some promising news!... I am order 1624X and I received my seek thermal camera on Saturday November 29th. Some backstory... I ordered on November 13th, received label created notification on November 17th, and I received this email on 11/25/14 after asking for the status of my order and received this email:

Ryan (Seek Thermal)
Nov 25 13:03

Hi Mitul.

There was an issue at the manufacturer with some of the shipping. Your order actually left the warehouse yesterday with a new label and tracking.

XXXXXXXXXXXX

Sorry for the confusion.

Ryan
Seek Support

[-jeffB]

That IS encouraging news! My order was slightly behind yours, and I'm hoping I have the same issue (dead tracking number). I'm a bit alarmed, though, that I've received no response to my emails. It's not giving me a warm fuzzy feeling about the company.

Thanks for the response! I'm hoping to perhaps dust off my ImageJ skills once I have the unit in-hand...

[eneuro]

Argh, cant get the square brackets to escape, still trying.

Maybe Non-Local Means Denoising might be usefull with thermal images too 
Tried this  online denoise demo app where you can upload image to denoise, but not great results with preprocessed Seek thermal raw data, and for the moment Flir's E4+ images looks better.
However, need to improve also automatic debanding and for the moment in recent project in thermal objects detection by security system simply median blur worked fine 

Will try my own custom version of this local means denoising algorithm while those built into OpenCV has limits for 8bit channels only and more time needed to tweak it to work fine with Seek thermal images.
It is time to display some real temperatures and crack other information hidden in Seek hexagon patent pixels.
Anyway no problem to record thermal videos from this dongle using custom 16bit iron LUT, so it is fun to play with images from this camera in own OpenCV software, detect ghosts and thermal objects at night without any visual light source  and record video to review later

[miguelvp]

I meant the square brackets on the link that I wanted to share, Thanks Giving weekend I didn't get to do a thing and at work I'm on crunch mode, so no image processing for me for a while.

[Marco]

NLM is too clever by half ... way too much chance of introducing spurious detail. Just stick to masking of known broken pixels and something simple, like bilateral filtering.

[eneuro]

NLM is too clever by half ... way too much chance of introducing spurious detail.

NLM can be also computationaly too expensive on some less powerfull machines, so after looking into this nice Bilateral Filtering by Eric Yuan's Blog comparision of other popular filters vs bilateral filtering it looks like it should be fine and test visual images at the end of this article are very nice looking 
Lets map a few keys to this filter parameters and see how it performs with Seek thermal images, but in OpenCV we can even control border type during this filtering and it is one line of code and can be used with float and even 3 channel images, so the most important thing now is tune bilateral filter parameters to fit into Seek thermal images size and its values range.
I have a few basic statistics of calibration and image frame raw sensor values, as well as output thermal image  like its mean value and standard deviation, so it should be easy to find proper parameters for bilateral filter based on needed output resolution and specific thermal image processing task   

[WarHorse]

Hi! I receive my Seek camera... After unpacking I saw small traces of 'glue' or 'compound' UNDER front lens. It's can be find only with magnifier - very small and masked with front lens glazing...
Is anybody have that?
What is that? Bug in assemble or it normal? Is it will not work properly?

And sec quest: My is micro-usb version. Is it can be usable with lightning apple devices? Through any adapters or maybe other methods?

[-jeffB]

Well, apparently my camera was delivered on Saturday, and has been sitting in a corner of our porch ever since.  Wrong tracking number, and never a single response from Seek customer service. Not the way to cultivate success, guys.

Having said all that, the camera itself is a blast! Yes, I see the gradient issue; yes, it's noisy; yes, the auto-ranging is often annoying; yes, I have to use an extension cable because my device's USB port is oriented "the wrong way". Don't care; I'm having fun.

Seek's site notes that the 2012 Nexus 7 doesn't work with this dongle, but our 2013 model worked the first time. I haven't tried it yet on my wife's Nexus 5, and it certainly won't work on my vintage LG Marquee.

Squirrels had chewed through the soffit under the roof above our bedroom, and when I pointed the Seek at the ceiling corner of our bedroom, this is what I saw. I went and thumped on the hot spots, and heard scrambling noises from above. Time to call in pest control and a carpenter, before those hot spots start to show cold (wet) spots...!

[frenky]

Seek has put old app back on Play store (1.4.0.2):
https://play.google.com/store/apps/details?id=com.tyriansystems.SeekThermal

[efahrenholz]

Wow. Someone must have really borked the software if they can't figure it out. I'd say they have had enough time to find the errors...

[Uho]

Conducted several experiments with the module. Use the program for is windows. 1. Very large gradient. 2. Large thermal noise. 3. External lenses only work for macro photo. When installing telephoto sensor sensitivity is low. Noise obscures.
That's all we know. I just confirmed. Program for is windows has disadvantages. Perhaps the author if time will want to correct them. Orientation is not correct. And if you deploy disappear inscriptions. Installed on Windows 7. The resolution of 1024x600.

[eneuro]

External lenses only work for macro photo. When installing telephoto sensor sensitivity is low. Noise obscures.

I'm not sure which filters are used in your app and in oryginal Seek Thermal android app-not too many options available in this software-anyway just implemented in my Linux thermal interface custom filters for huge spikes filtering and bilateral-bilateral filters works fine, but it had problem with this very noisy frame below (only a few Seek raw sensor frames and banding not removed in this test-at the end only low temperatures manualy tuned), so implemented additional much stronger filer before bilateral which removes those large noisy pixels and at the end result might be close to Flir's One smoothed image for their MSX 


While many different filtering methods are already implemented in my open source pre release version of thermal interface it is time to play more with maybe additional hidden hexagon patent pixels values and examine those mysterious first frames when Seek device is powered on 

BTW: If you have @Uho option in this Win PC app to export some raw sensor data and publish it could be interesting examine another startup sequence of another Seek dongle, while collecting different Seek dongles raw sample data to guess what is comming from different Seek sensors at different conditions, so ambient temperature and how long this device was turned on could be helpfull for anyone interested in improvements of any Seek unofficial developer tools code 
Just adding to my software tools for analysis of dead pixels paterns in Seek Thermnal dongles to see if it varies somehow in temperature between devices which could affect also output image quality, but it looks like lenses and shutter constructuons are main limits of this quite interesting dongle, but by adding additional visual light camera and merging this thermal and visual images with applied gradient correction we should get very nice looking imaginery from this device including stereo thermal images while two such cheap Seek dongles can be used 

[eneuro]

Don't care; I'm having fun.

It is more fun with 16bit thermal LUT tables  like this iron -those Seek images looks very noisy and since there is no sharp edges maybe it is better smooth flat areas a little bit while image looks much better and applying 16bit thermal LUTs makes those images more friendly to human eye, so it should be option in Seek software to adjust smooth and noise reduction levels manualy when needed as I've did in my thermal interface 


Anyway it is interesting why Seek do not provide for the moment any thermal LUT scale like in Flir's cameras, while it is a must to have feature in such software 

Will see when Seek will add this to his official app-can't understand why it is not already done.

Today implementing such thermal interface OSD, to be able adjust thermal LUT temperature ranges and center pointer cursors, so looking for any good Flir's or SKF'a user manuals to see how it is already done and improve it.

BTW: Jpeg 100% quality and PNG compression level 9 with 16bit LUTs looks no so bad, so quick switch between those JPG100/PNG9/GRAY16bit output images seams to be quite usefull save feature and should be fine for the moment to play with this dongle and publish thermal images taken from this device including video recording in 16bit LUT RGB mode.

Edit: Those SKF TKTI 21 & 31 users manual has realy nice detailed menu options explained 
For example it has a few cursors with different emissivity setting for more accurate temperature readings which is a feature what we must to have in our custom Seek Thermal interface 

Nice sharp thermal images from this SKF TKTI 160x120 (FPA) thermal camera and powerfull PC software for analysis 

[Uho]

The program creates two files. TXT and DAT. DAT is greater than 1 megabyte .

[tomas123]

@eneuro: I see my bad pixel image 

if you interested in mapping files of Ex /Exx see this post
https://www.eevblog.com/forum/testgear/flir-e4-thermal-imaging-camera-teardown/msg345231/#msg345231

and the following posts with the attachments

mrflibble made in December 2013 some nice researches about correction images
https://www.eevblog.com/forum/testgear/flir-e4-thermal-imaging-camera-teardown/msg343483/#msg343483
https://www.eevblog.com/forum/testgear/flir-e4-thermal-imaging-camera-teardown/msg345296/#msg345296
https://www.eevblog.com/forum/testgear/flir-e4-thermal-imaging-camera-teardown/msg345521/#msg345521

every Flir Ex/Exx save an image from black body (calibration)

see
http://www.flir.com/cs/emea/en/view/?id=41491

[miguelvp]

The program creates two files. TXT and DAT. DAT is greater than 1 megabyte .

Select the .dat (or both) right click and select "Send to->Compressed (zipped) folder" it should be shy of 1 megabyte.

[Uho]

I tried to zip. Does not compress. Succeeded in rar.

[Rasz]

Anyway it is interesting why Seek do not provide for the moment any thermal LUT scale like in Flir's cameras, while it is a must to have feature in such software 
Will see when Seek will add this to his official app-can't understand why it is not already done.

cant be patents, afaik part of Seek/Flir settlement was access to flir patent pool

[mikeselectricstuff]

Anyway it is interesting why Seek do not provide for the moment any thermal LUT scale like in Flir's cameras, while it is a must to have feature in such software 
Will see when Seek will add this to his official app-can't understand why it is not already done.

cant be patents, afaik part of Seek/Flir settlement was access to flir patent pool

Almost certainly not all of them though

[efahrenholz]

They haven't added a scale because this camera wasn't aimed at people who needed an accurate camera. This is like a point and shoot disposable camera competing with big dslr's. We have a fair image to work with; it's passable for quality and resolution (like a Polaroid), but there's only so many practical things to expect from this setup. Adding a temperature scale to the output image is like nutrional facts on the lid of your Pringles can. Its ultimately in the way. They want the chips. 

[Nik]

Image palate shifts while seeking. On a low dynamic range scene the background mostly yellow. Then if put a high/low temp object (eg finger/ice), main background become red/black therefore it has the same temp.
Or image post-processed and become non linear at all.

[eneuro]

They haven't added a scale because this camera wasn't aimed at people who needed an accurate camera.

Even if it will not be great accuracy without temperature LUT range in OSD we have no idea which objects in the scene are at room ambient temperature range and this what we see is complete mess, so while it is only a few lines of code to create choosen thermal LUT preview and assign it with choosen temp limits etc and display it as an option like in many thermal cameras of course I won't wait for Seek and will have it done today I hope, since have idea how it could be done to be very easy to use on touchscreen devices and classic PC machines with mouse pointers 

@Uho   Thx for those startup frames from your device - I will add it to collection of  other peoples data files and create by my automated tools map with dead pixels taken from a few calibration frames marked as read and hexagon patent Seek pixels marked as green on those raw sensor 208x156 pixels 16bit files 

@tomas123 Yep, it was very interesting to see Flir's dead pixels patterns and will look into its calibration frame image to see how those dead pixels are distributed, I mean how many of them are connected together forming bigger dead  pixels group.

It is interesting if manufacturers of those microbolometers have any quality measures and limits for example for maximum number of such connected dead pixels which can be catched on those calibration frames?
It looks like different devices can have very different those dead pixels patterns and there are devices where they can can form even 4 dead pixels close together which creates one big dead pixel and it passed any quality tests? I guess there is not only percentage of dead pixels per device as quality measure 

[-jeffB]

I understand perfectly well why Seek (and FLIR and everybody else) default to auto-ranging when rendering images. Heck, even ImageJ did it by default, and we had a bit of a struggle to get around it when working with radiological images.

But I really, really want a tool to lock down my display range, and adjust that range manually. I understand and accept that it will mean much of my scene may go full-black or full-white. But when I'm looking for critters in the woods, I don't want a glimpse of sky to turn all the trees near-white, and when I'm checking interior walls, I don't want a glimpse of an incandescent bulb to turn everything else dark-blue.

It would be great if Seek would provide a "power user UI" with a few features like this, but I'm pretty much figuring I'll need to write it myself, and given my past performance on such projects, I'm not holding my breath. All the same, I am following the work discussed here with great interest...

[eneuro]

Succeeded in rar.

I suggest probably better use this @cynafab python code to catch raw USB Seek images, while it looks like this data file from @ miguelvp in version you might have IS NOT raw sensor data, but preprocessed somehow by @miguelvp 

Just created bad dots for reliable and examined by us @cynafab python code from this his post Here is a great way to get started investigating the Seek Thermal Camera. by cynfab and images provided here taken from this software there Sample raw data by cynafab and I've got as expected a few bad pixels (23) on calibration frames 0x1 and 0x3 marked in this gray scale 8bit map output with white 255 values (128 has those hexagon perfect patent Seek pixels):

Code: [Select]

../img2pat: Seek Thermal sensor hexagon pattern image dimensions: 206 x 156 value min: 128
../img2pat: Seek Thermal raw sensor image patterns written to: pat.rawData11_6.png  bad pattern pixels: 23/32136  (0.072 %)

../img2pat: Seek Thermal sensor hexagon pattern image dimensions: 206 x 156 value min: 128
../img2pat: Seek Thermal raw sensor image patterns written to: pat.rawData12_1.png  bad pattern pixels: 23/32136  (0.072 %)
and they are attached to this post:


Similar results I've got with @marshallh calibration frame which he provided here and it looks like this (28 bad dots):

They are slightly different distributed, but it is clear there are some bad dots outside those heagon patern patent pixels.

When taken data provided by you @Uho few posts above... I've got close to perfect sensor as well as in the case of @miguelvp which means that probably you HAVE NOT raw sensor data anymore from latest @miguelvp software 


@miguelvp Can you confirm that you allow to save only preprocessed Seek images from your latest software while my guess is you save in data.dat file output those frames with ALREADY filled bad dots somehow, so we got only a few 5-6 bad looking dots in the case of this output calibration frames and this data is useless if someone wants to see how many bad dots has his Seek dongle?