Actual resolution of Flir One V2

[tom66]

Does anyone know the actual resolution of the 2nd gen Flir One which supports Android?

Flir seem to indicate 160x120, but they say this is an "enhanced" resolution, and it uses a Lepton core (which seems to be 80x60 only)

I'm tempted to get one and sell my Seek Thermal, but I think 80x60 even with MSX will be too low resolution for SMD work. I plan to add a lens to it, to use it with SMD, so MSX would usually be off.

[TopLoser]

Got one on my desk. What's the simplest way to determine actual resolution?

[tom66]

Look at an image with sharp gradients, you will be able to tell if it uses interpolation or something. Just upload a photo here and we can probably tell.

[TopLoser]

Just took a pic of a finned heatsink. MSX turned off.

[tom66]

Hard to pick out any contrasting edges - could you get a picture of something with a sharp edge? Maybe something like a soldering iron, or a thin and hot wire.

[TopLoser]

Well you can easily resolve 30 fins. That's surely more than 60 pixels in the horizontal direction?

[tom66]

Yes - I appreciate that and thanks for taking the photo with the camera. But I am trying to determine if the resolution is greater than 80 pixels across. That sets a lower bound of 60 pixels but we know the Flir Lepton sensor is at least 80x60. I'm just wondering if Flir have a custom 160x120 Lepton, or if they're just talking about some edge enhancement algorithm that improves apparent sharpness but doesn't actually add extra data.

It does look remarkably better than my Seek, and at a price that is extremely competitive.

[TopLoser]

It was a rushed picture, I have 'stuff' to attend to this afternoon and over the weekend. I'll try pictures with an increasing number of fins when I get back. I also have a Testo camera with 160x120 resolution I can do the same tests with.

Post of pic from the Seek camera with 30 similar fins if you can please? Or if somebody has an existing Flir One they can try the same.

[Fraser]

I have written to my friends in FLIR on this topic and hope to have a definitive answer soon.

Aurora

[frenky]

Post of pic from the Seek camera with 30 similar fins if you can please? Or if somebody has an existing Flir One they can try the same.

I don't have anything similar at hand but this photo shows what Seek is capable of:
(With custom software: https://www.eevblog.com/forum/chat/seekthermal-how-to-correct-the-image-received-from-the-sensor/)


I'll try to find a pcb that has even smaller pins on IC.

[Trax]

whats the FOV of the new flir one?
can someone make a picture of a PCB?

@frenky
Have you used any additional lenses?

[frenky]

Yes I used 50mm lens in front of seek.

Sent from my LG-D855 using Tapatalk

[Muxr]

So I did a little experiment. I took the original image, and downscaled it to both purported resolutions, just to see how much data is lost. I don't know that this is a valid test. Since image resizing algorithm is obviously going to discard more when squeezing the image into a smaller resolution. But I still think it's interesting.

edit: also there are some jpeg compression artifacts at play, but I used 100% quality in every step.

[NathanFowler]

Not sure if this helps but here is an ambient temp heat sink against my hand.  Seems we can see the fins, gen 1 max blending off

[Fraser]

Well executed interpolation can be challenging to spot on thermal images when looking at everyday objects. The 160x120 image is up-scaled to the resolution of the phone screen as well.

A side by side comparison with a reference 160x120 pixel camera image is likely a good way to spot any obvious image manipulation. A common target, such as an SOIC chip, could be used to compare edge detail around the pins.  I can produce 160x120 images from an NEC AVIO F30 or TESTO 875. I will see what I can do over the weekend.

There are some clever image experts on this forum. Maybe they know how best to spot multiple levels of interpolation.

Aurora

[Muxr]

Ok so I added a 3rd sample. In it I tried to exaggerate the loss due to recompression and bicubic scaling. It's the original image, downscaled to 160x120, then upscaled to 320x240, and back down to 160x120 again.

You can tell that the image maintains the original detail. Where as going down to 80x60 has produced a significant loss in detail. My feeling is that the new camera indeed has more resolution.

[NathanFowler]

What's strange is the Moiré pattern that's so pronounced, is this usual?  It persists on all samples in a very clearly defined pattern.  I defer to your expertise but does this point to post-processing issues on-device or interpolation?  Very interesting and thank you.

[Muxr]

I don't know much about how thermal imaging sensors work, but the moire pattern is common in digital imaging. Most imaging color sensors use a Bayer pattern and they all require an anti aliasing filter to combat moire. Some super high resolution camera makers do away with the filter in order to restore some of the sharpness lost due to the AA filter (we're talking 36mp +). Fujifilm uses X-trans sensors which combat this by laying RGB fotosites in a different pattern.


You can see the difference in uniformity. of RGB photosites.

This is why film didn't have moire, because it had random distribution of pigment.

So moire is to be expected on sharp edges. I doubt you'd want an AA filter on a thermal imaging sensor. Also we're still dealing with really low resolutions here where accuracy is more important than the pleasing look.

[NathanFowler]

Muxr thank you for taking the time to explain, I apologize and now realize you were taking a singular posted image and applying resolution degredation as an approxmiation for resolution differences.  I would assume the same Moiré pattern as you upscaled and downscaled the same image.  Apologies on my part; I saw the patterning and wasnt' sure of it's origin but you've explained it very well and I've learned something new today.  A sincere thanks.

[Muxr]

Muxr thank you for taking the time to explain, I apologize and now realize you were taking a singular posted image and applying resolution degredation as an approxmiation for resolution differences.  I would assume the same Moiré pattern as you upscaled and downscaled the same image.  Apologies on my part; I saw the patterning and wasnt' sure of it's origin but you've explained it very well.

Exactly, my crude method was really an experiment to see if an 80x60 image could contain the detail seen in the original image. After comparing the detail lost when resampling down to 80x60, I think not. I think the new sensor does indeed have a higher resolution.

Sorry for the confusion.

[NathanFowler]

Sorry for the confusion.

It's I that should apologize and thank you for the explanation and taking the time to do so.  To use an American colloquism, it just "clicked" meaning all pieces of the mental puzzle suddenly were in place.  I very much apprecaite your explanation and taking the time to educate me.

[jadew]

A common target, such as an SOIC chip, could be used to compare edge detail around the pins.

I can help with images from my Seek. It's the XR version so I didn't use additional lens.

The square IC in the first one is a QFN32, so 0.5 mm pitch. It's also present in the second image.
The chip in the second image (bottom left) has a 0.8 mm pitch.

The dimension of the first image is about 38 x 28 mm, while the second one is 11 x 8 cm.

I also attached enlargements of both with bicubic interpolation.

[tomas123]

There are some clever image experts on this forum. Maybe they know how best to spot multiple levels of interpolation.

Aurora

see here for sample images from a heating wire and some ImageMagick steps
https://www.eevblog.com/forum/testgear/flir-one-thermal-imaging-camera-teardown-and-hacks/msg551882/#msg551882

[TopLoser]

I don't have any lenses of any kind and the Flir is fixed focus so there's no point me trying to take photos of boards. So don't ask.

After I took it apart and bodged it back together I put the heatsink back in the oven again and tried a few pics at different distances to see where it became impossible to resolve individual fins.

With the Testo 875-1i (manually adjustable focus) this is the furthest I could get before it all became a blur:



A quick measure of the stepped jaggies easily reveals a resolution of 160H x 120V. Exactly as claimed by Testo for that model. There's no significant attempt at interpolation or cheaty processing that I can see in that pic.

On to the new Flir. Furthest I could get and still see all the fins:



Lots of interpolation and processing going on, no obvious jaggies and it all just rapidly turned into a blob if I moved a few cm more away. Counting horizonontal fins I guess it's *just* possible to resolve about 50 across the entire width (that heastsink has 30 and it doesn't fill the entire width). Double that up and we have 100, not quite the 120H we were looking for but not far off considering the focus is probably crap at the distance the picture was taken. Sorry I didn't rotate the camera 90 degrees to match the Testo pic...

I'm not a fanboy, I'm not trying to sell these things, in fact I have no need for them. I'm just curious. But if somebody gave me one of these and asked what the sensor resolution was then right now I'd say it was 120x160 considering the poor focus at short distances. I'd rather Flir didn't f*ck about with interpolation and processing because it doesn't add anything in my view, the raw images would perfectly acceptable combined with MSX.


Edited to get some details right lol

[Fraser]

For anyone not familiar with what IRISYS did to enhance their low resolution camera images, I attach an image showing what they achieved.

IRISYS were working with a low resolution 16x16 pixel pyrometer chip that needed a lot of interpolation to achieve a useful image.They interpolated from 16x16 up to 128x128 ! As many readers will know, interpolation adds no true data to an image, it just takes the values of adjacent pixels and inserts virtual data in the midpoint between the two. Depending on how well interpolation is implemented, it can be challenging to spot in a processed image, especially if other image processing algorithms to remove sharp transitions are in use.

In a visible light camera you use a resolution chart to see the true imager resolution. This gets below any interpolation that may be in play. Sadly I have no such chart for use in the thermal domain.

A thought to consider.......

FLIR may be using sophisticated interpolation, and other image processing techniques, that may challenge us to determine whether the microbolometer is native 80x60 or 160x120........ If this is the case, they have done a very good job  If they are operating a X2 interpolation and we cannot truly see such in the real world, we are unlikely to have issues with the images that the camera produces in daily life 

If our tasking is so demanding that it needs 160x120 physical resolution, then we should be able to clearly see the image degradation when compared with a known 160x120 thermal camera.

If not.....be happy 

Aurora