How do you binary search for two people arriving, one punches the other, they both leave?
In the same way the OP talks about it …
You don’t watch the whole thing, he said. You use a binary search. You fast forward to halfway, see if the bike is there and, if it is, zoom to three quarters of the way through. But if it wasn’t there at the halfway mark, you rewind to a quarter of the way though. Its very quick. In fact, he had pointed out, if the CCTV footage stretched back to the dawn of humanity it would probably have taken an hour to find the moment of theft.
Instead of a bike, you look for the aftereffects of a fight happening (chairs knocked down, tables turned over, etc.). You can even look at how many people congregate around the location of the fight before and after the video as a ‘marker’ to the point of time the fight was happening/just finished.
Edit: One thing we didn’t even mention, AI can also be used these days to notice subtle changes in the video. If a video is a static image of an alley, then two people walk in the alley and fight, even though they leave no traces behind, that movment of the fight is caught on the video. Motion sensor movement, basically.