Dark frame subtraction

[russss]

Here's a proof of concept for doing dark frame subtraction. It takes an image before turning on the LED, then subtracts that image from the one taken when the LED is on.

This results in an extremely clean image. The only disadvantage is that it's very sensitive to anything moving in frame - even sub-pixel movements can become visible.

This managed to accurately detect LEDs in a large room with my (old) Macbook Pro webcam and no exposure adjustments. I think exposure adjustment may still be valuable, but this makes it much less necessary.

This might not be the neatest way of doing it, as I hacked a lot of stuff around on Friday. I've refactored this a bit without testing because I don't actually have any easily-accessible LEDs at home...

I'd recommend giving it a try with your test setup and see how it works.

ref #33

[TheMariday]

I've had a play with this this morning and it operates as I expect!

Howeverrrrr, this does raise some other things that (I) need to do:

• Update documentation so camera exposure is less prominent
• Add an escape clause for waiting for darkness in case the camera moves slightly and a large swathe of light area appears
• Update the camera_check tool so it performs in a similar way to the detector now and document

Oh only last thing is that cv2.absdiff will return a high value if an area of the image gets darker as well as lighter.
So I've replaced that with cv2.subtract in my copy