Detecting frame changes efficiently

[Bryan McCormick]

I have to cobble something together which is not normally my thing, a 
video app that doesn't do a bunch. What it does need to do though is 
detect when there has been a significant change in frame.

I would need to be able to threshold this for significance, so I would 
need to know that xyz percentage of the frame has changed. I might not 
want the app to trigger unless there has been a 90 percent change, 
something like that though I am not sure yet how high that rate is going 
to have to be. I will probably sample smaller frame elements to try to 
keep it lighter on overhead.

My approach was to do this not with 2 frames but with 3 sampled over a 
few seconds so that there would be a higher confidence that the change 
isn't just frame a lighting change or something that is transient.

I already can do bit by bit stuff, but that is super super slow. I was 
using RGB compares.

Does anyone have any ideas about how to do it faster? I was hoping there 
might be some method to do it by rows in a still image grabbed from the 
stream.

Importantly, I am grabbing the images off as stills and storing them, 
then overwriting as the sample time window passes.

Thoughts?