Identify "outlier" frames in video

[sfmig]

From a chat with Dario and @niksirbi, it would be very useful for researchers to automatically detect "outlier" sections in a video
(E.g., when the experimenter puts the animal in the arena or a human walking in)

Later we were wondering if there could be a simple method to do this, maybe focusing on static cameras to start with.

Could detecting a large difference with a mean / median frame be enough?