Improving thermal sensor resolution by vibrating/moving sensor while shooting

[bap2703]

The depth of field is too shallow so it's hard to compare because the focusing dominates the blur


Superresolution should work better at hyperfocal distance than for macro work.

[SilverSolder]

Very cool set of tests!

With a 640x480 native sensor, it will probably move by one pixel between each shot just sitting on a tripod...  e.g. with a 36 megapixel camera, it is totally impossible to shoot the exact same set of pixels twice, the camera just has to move by a tiny fraction of a bee's d!ck...        

Even with a high resolution DSLR, you can benefit from making a focus stack (i.e. you aren't trying to fix a bad sensor, you are trying to get a deeper depth of field).

Looking at the image in the original post, it seems to illustrate that with a very low resolution sensor like the Seek, the "manual superresolution" adds a lot more value than it does to a higher quality thermal camera.  (At the cost of adding a lot of time and work to the process.)

[Ultrapurple]

@bap2703 - yes, you're right, hyperfocal would be a lot better. I simply didn't think of that when I was sitting at my desk tinkering with the Pi! I'll do something outside when I have a spare moment. I shall be intrigued to discover whether I can produce the 'same' image shot-to-shot on a tripod (plus or minus image noise).

Agreed on the focus stacking. I have an old camera that has this built in (it isn't terribly good) but a friend who does a lot of product shots has a modern mirrorless camera that does in-body focus stacking and is stunningly good.

[SilverSolder]

[...]

Agreed on the focus stacking. I have an old camera that has this built in (it isn't terribly good) but a friend who does a lot of product shots has a modern mirrorless camera that does in-body focus stacking and is stunningly good.

I like the in-body focus stacking idea.  Might eventually make it into a thermal imager near you? 

[frogg]

I think there's a pretty long practical history of this approach working with raster imaging on FPAs across a broad variety of wavelengths...there's definitely no reason why this wouldn't work with LWIR.

I think the real challenge is to create a method that would create accurate radiometric high resolution images.

[SilverSolder]

I think there's a pretty long practical history of this approach working with raster imaging on FPAs across a broad variety of wavelengths...there's definitely no reason why this wouldn't work with LWIR.

I think the real challenge is to create a method that would create accurate radiometric high resolution images.

Sometimes photographers put a color chart somewhere in the scene, to get accurate colors.  How about putting a device with a known temperature (or two or more known temperatures, to give a range) in the 'thermal scene' to act as a reference for a radiometric image?  That way the imager doesn't have to be perfect.

[bap2703]

I think the real challenge is to create a method that would create accurate radiometric high resolution images.

You mean from the software tools available to hobbyists?

[frogg]

You mean from the software tools available to hobbyists?

Well, I mean a freely available closed form solution (mathematical and/or algorithmic) with associated source code, plus supporting hardware.

I guess if I am a hobbyist with deep pockets, I could just buy Ultramax. But that's not what I want. I don't just want to buy some dude's product. I want all the source code, theory, and knowledge.