Yet another cheap thermal imager incoming.. Seek Thermal

[Fraser]

@Mike,

That will be very sad indeed because the camera appears to have an issue that will win SEEK Thermal no friends if they show no interest. As we know the user community will be quick to look for alternatives if this is nothing more than a toy with 'issues' and no support. A great pity if this comes to pass. There is that other thermal camera project that tried to get Kickstarter backing a while back. I believe it uses a German sourced detector array.

I have not given up on the SEEK device just yet though and if we can discover the cause of the issue, then DIY correction is a possibility. BUT at $200 I am not that interested in spending tens of valuable man hours sorting out SEEK Thermal's problems for them if they don't show any interest in supporting their product. I would just buy another thermal camera and move on..... except I don't need any more thermal cameras 

Aurora

[miguelvp]

That microbolometer still looks hotter than I would have expected but interestingly there is no evidence of a hot spot on the rear of the PCB under the microbolometer.

It will be very interesting to hear SEEKs thoughts on this matter, hopefully tomorrow.

Aurora

The PCB is pretty thick there there is probably a ground plane within it's layers.
Seems like they designed it that thick to avoid thermal issues.

[miguelvp]

They do have a twitter account, maybe that would give a faster response if they don't answer via e-mail? I would hate to be too quick to bring it up on twitter and give them some time to address the issue privately.

[eneuro]

After being on for a while and the shutter forced open if I force it close, the image is less even but not terrible:

Which LUT are you using? Is it possible in Seek Thermal app disable any LUTs and get grey output?

Thanks to @marshallh he provided another ones raw sensor USB captured images 
Played a while ago with those images-disabled any gamma corections in my software and set smooth to level 9 so more averaging, than applied iron256 LUT mentioned before (2nd row coloured ones)

After that I've converted them to gray again and normalized and... I do not know those images from left to right looks more uniform when we blur them a little bit (average a few times) and normalize or equalize, even using any purpose image software like Gimp starting with this raw sensor images provided by @marshallh
I've included them in last row, with left to right timeing like other above starting from oryginal in the first row...
I do not resized them from its oryginal 208x156 to do not create additional artificial pixels eg. by linear resize 4x to Seek Thermal output resolution 832x624, so they were processed at its oryginal sensor size 208x156.

It is interesting what explanation might be and if it is hardware issue 

BTW: Still those last two columns in raw sensor data are mistery-simply skipped them in my image processing and shifted whole image by 1 pixel to nice horizontal fit into 208x156 sensor raw data image size.

[mikeselectricstuff]

Bear in mind that (AFAICS) they don't give any  spec whatsoever for absolute accuracy or NETD, and there are inevitable compromises making something so cheap. For the majority of their target market, it's good enough.
 
There is no temp sensor on the shutter - it appears that the only temperature sensor is a diode on the rear of the PCB, so is it reasonable to expect better than 5-10 deg.c accuracy?

These folks are not newbies in the thermal imaging industry, and it's quite possible that the performance is as good as they can get for the price.  What would be interesting is to compare a few units to see how similar the issue is.
They may well be using the phone add-on market to hone the process to a level more acceptable for higher end users, for the next product. Flir are already marketing Lepton for industrial monitoring type apps, but I think the Seek is at least one iteration away from being able to compete in this market.

[eneuro]

The shutter arm may also conduct away heat from that part of the flag.

Talking about DIY shutter I was thinking about one... without any physical connection with PCB 

[miguelvp]

Leaving the shutter open so it references whatever is looking at and moving the camera slightly to the righ or left or up or down does a pretty cool edge detection, Of course a regular camera will do the same

On the blob front I see Mike's point on not necesarelly being the shutter since if I leave the thing open and steady to any background with whatever number of heat differences when it subtracts the reference image it will always look flat (or if I didn't move the camera when I took this picture)

It's on my wife's phone and it's setup as a 16:9 ratio because it's easier to find things on that mode.

The picture is a chair with a window with blind behind. A humidifier on top of the chair (not turned on) and the edge of a radiator on the right.

[marshallh]

I put my finger next to the shutter by accident... still far from the sensor, and I see a nice white blob along the left side of the screen. Sensitive to indirect heat even thats a few mm perpendicular...

[miguelvp]

So as the chip gets hot the temperature difference with the window from where the shutter is at is the reason for what I call the blob?

It's 6 degrees difference more or less in my case

[ebeall]

@Aurora,
There is indeed another imager coming soon that had been developed using a German array. It would be nice to use the Seek but the performance is not good enough now. The noise is a problem, but its still early, they may have better performance in a different configuration or future version, as you guys are hard at work on figuring out.

[miguelvp]

Marshallh, does that seeker actually captures data directly form the PC or you are capturing the data first then using that to decode it?

[marshallh]

It is a live preview, controls the device directly (I don't have a smartphone of my own)

[sgstair]

Seeker app was linked previously in the thread also, it's available here: https://github.com/sgstair/winusbdotnet
I still have some plans to improve it, at the moment it's just doing very dumb calibration on the raw data and rendering it.

[miguelvp]

Awesome, thanks sgstair. I do have a micro b to A adaptor in order but I took the cheapest one so it's going to be a while until I get it. I guess I will have to sacrifice my micro B extension cable in the meanwhile. or desolder a female from an old phone.

Cool will download it asap, thanks again.

[marshallh]

New app version was released on the 24th:
https://play.google.com/store/apps/details?id=com.tyriansystems.SeekThermal

What's new:
* Corrected crashing issues on various devices.
* Image quality improvements.
* Other misc. Improvements.


Can someone confirm if it has a firmware upgrade?

[miguelvp]

I did sacrifice the micro B USB extender data cable:


Updated the winusb driver with Zadig as previously posted (maybe that should be part of the readme )
Compiled it with visual studio express 2013 and was a piece of cake:

My hot router

Thank you sgstair!

@marshallh I did have to download the apk with bluestacks to begin with so I don't get updates. I guess I can download the current one.

[miguelvp]

New app version was released on the 24th:
https://play.google.com/store/apps/details?id=com.tyriansystems.SeekThermal

What's new:
* Corrected crashing issues on various devices.
* Image quality improvements.
* Other misc. Improvements.


Can someone confirm if it has a firmware upgrade?

Haven't checked but did a backup of the apk and placed it where I placed the last one:

http://www.miguelvp.com/ForMike/

Edit: Of course is the one named Seek_Thermal-v1.3.0.1.apk

Now I'm trying to see if I can make BlueStacks use the usb camera directly, Not sure if it's possible

[marshallh]

Image has new firmware built Oct 21.

[miguelvp]

Image has new firmware built Oct 21.

Cool, plugged it into my wife's cell to update the camera and when I ran the application it did mention updating firmware to your camera.

The image on the left is the new firmware, the one on the right is the previous one.

Not too scientific but I tried to get it as close as the previous shot.

[miguelvp]

So I started to play with sgstair's code, and in Form1.cs I added a bilinear interpolation, kind of crude but it works.
So for dead pixels it averages from the top, bottom, left and right, didn't bother to do the corner ones.

Code: [Select]

#if AVERAGE
                int[] deadpixelarray = { 10, 6, 2, 13, 9, 5, 1, 12, 8, 4, 0, 11, 7, 3, 14 };
                int deadpixel;
                ulong average;
                ulong numpoints;
                int v;
                for (y = 0; y < 156; y++)
                {
                    deadpixel = deadpixelarray[y % 15];
                    for (int x = 0; x < 208; x++)
                    {
                        if (x == deadpixel)
                        {
                            average = 0UL;
                            numpoints = 0;
                            if (x > 0)
                            {
                                average += data.PixelData[c - 1];
                                numpoints++;
                            }
                            if (x < 208)
                            {
                                average += data.PixelData[c + 1];
                                numpoints++;
                            }
                            if (y > 0)
                            {
                                average += data.PixelData[c - 208];
                                numpoints++;
                            }
                            if (y < 155)
                            {
                                average += data.PixelData[c + 208];
                                numpoints++;
                            }
                            if (numpoints > 0)
                                average = average / numpoints;
                            else
                                average = data.MinValue;
                            v = (int)average;
                            deadpixel += 15;
                        }
                        else
                        {
                            v = data.PixelData[c];
                        }
                        v = (v - data.MinValue) * 255 / (data.MaxValue - data.MinValue);

                        if (v < 0) v = 0;
                        if (v > 255) v = 255;
                        bmp.SetPixel(x, y, Color.FromArgb(v, v, v));
                        c++;
                    }
                }
#else
                for (y = 0; y < 156; y++)
                {
                    for (int x = 0; x < 208; x++)
                    {
                        int v = data.PixelData[c++];

                        v = (v - data.MinValue) * 255 / (data.MaxValue - data.MinValue);
                        if (v < 0) v = 0;
                        if (v > 255) v = 255;

                        bmp.SetPixel(x, y, Color.FromArgb(v, v, v));
                    }
                }
#endif
Edit: got rid off an unused variable and defaulted the dead pixel to min value instead of max value if it couldn't average (but that should never happen anyways)

Next I might characterize the background better since there is some patterns that can be adjusted it seems.

I added an Iron palette later on with paintshop pro.

[eneuro]

https://github.com/sgstair/winusbdotnet
I still have some plans to improve it, at the moment it's just doing very dumb calibration on the raw data and rendering it.

So, those 3 images on screen shoots provided by @marshallh are NOT exactly raw sensor data but you modify its values, for example those non hexagon pixels somehow and change from 14bit to 8bit to be able to render it in RGBA mode?

However, does it work on standard PC USB 2.0 port with additional USB adapters or only on latest computers with some kid of USB OTG support?

[mikeselectricstuff]

New app version was released on the 24th:
https://play.google.com/store/apps/details?id=com.tyriansystems.SeekThermal

What's new:
* Corrected crashing issues on various devices.
* Image quality improvements.
* Other misc. Improvements.


Can someone confirm if it has a firmware upgrade?

..and upload an .apk...?

[miguelvp]

It works on a standard USB 2.0 port but you have to install the winusb driver with Zadig
http://zadig.akeo.ie/

On another note, brute force approach to add the Iron palette directly to the code, I know I could have just make it so it will read a file with the palette but meh. Notepad++ helped me

I'm attaching the modified Form1.cs with a built in Iron Palette

Edit: change the number of bitmaps back to 5 i changed it to only display 3

Code: [Select]

                if (bmpQueue.Count > 3) bmpQueue.Dequeue();
it was > 5

The result:


@mike apk is here http://www.miguelvp.com/ForMike/

[frenky]

I was able to remove most of the dead pixels with this simple (C#) code:

Code: [Select]

            //get image
            Bitmap sourceImg = new Bitmap(@"C:\temp\thermo\seek.png");
            Color leftPixel, centerPixel, rightPixel, topPixel, bottomPixel;
            int avgVal;

            //loop trough horiz pixels
            for (int x = 0; x < sourceImg.Width-2; x++)
            {
                for (int y = 0; y < sourceImg.Height-2; y++)
                {
                    leftPixel = sourceImg.GetPixel(x, y + 1);
                    rightPixel = sourceImg.GetPixel(x + 2, y + 1);
                    topPixel = sourceImg.GetPixel(x + 1, y);
                    bottomPixel = sourceImg.GetPixel(x + 1, y + 2);
                    centerPixel = sourceImg.GetPixel(x + 1, y + 1);
                    avgVal = (leftPixel.R + rightPixel.R + topPixel.R + bottomPixel.R) / 4;
                    if (Math.Abs(leftPixel.R - centerPixel.R) > 20) //if pixel is too different use average color from near pixels
                    {
                        sourceImg.SetPixel(x + 1, y + 1, Color.FromArgb(avgVal, avgVal, avgVal));
                    }
                }
            }

            sourceImg.Save(Application.StartupPath + "\\clear.png");
Before / after image:

[mikeselectricstuff]

I have managed to significantly reduce the gradient on mine by moving the lens position. Video soon.