You would have to be confident that said change in environment was done by the bike thief.

Well, the change would happen, the human will be noticed, and then they can watch that moment in time on the tape to see who did it. The binary search would be about shortening what portions of the video tape a human/AI would have to review manually.

It might be better to use some software to remove any frame of video that is identical to the one before it, no motion is taking place, etc. then manually watch the much shorter video of “only when stuff happens.”

So, I hope you’re not under the impression that I’m advocating binary search as the ONLY way of doing a search. I’m just staying within the confines of the subject as brought up by the OP, which was about binary searches.

At the end of the day its about detecting the change/aftereffect, and not the search inandof itself. A binary search just helps you narrow down the video you have to watch manually, especially when there’s allot of it to review.