Detect image without curves

[ethaniel]

Dear IM community, I've been banging my head against a wall for 6 months now.
I have a live beach camera which sometimes loses it's signal and produces a still image (which has no curves):


This is a example of a good image:


Is there non-expensive computing wise way of detecting the first image? (the camera is connected to a udoo board, which can barely pull the broadcast of the stream). The problem is that it always contains different colours (but the shape stays pretty much the same).

I am thinking that a good approach would be to count the number of curves in the image (but couldn't figure a way how to do it).

[snibgo]

How often is an image captured? A good method might be to compare adjacent frames: if they are similar, they are both good, but if they are very different then at least one is bad.

[ethaniel]

I tried comparing 2 frames, but sometimes there are variations in 2 "bad" frames which are higher than variations in 2 "good" night frames. So it's not a 100% optimal solution.

[snibgo]

That's what we want. A small difference between two good adjacent frames, and a large difference between one good and one bad, or two bad frames.

If two adjacent bad frames have a small difference, then the test needs to be more complex.