The research paper even has a cool sci-fi name: Eye in the Sky. The project uses an inexpensive to watch crowds of people from above. Then it uses AI to identify people in violent poses (the paper mentions strangling, punching, kicking, shooting and stabbing).
It's important to note that the researchers didn't fly a drone around to detect real violence -- The Verge reports it shot its own video clips using volunteers. So it's not clear how well it'd work in practice.
And even in an experimental state, the surveillance system raises questions around accuracy. The paper says it can spot a single person committing a violent act with pretty high accuracy, depending on the pose (89 percent accurate for punching, 94 percent for kicking, etc.). But overall accuracy also goes down as more violent people are added to the group -- it can spot one violent person in a group of 10 with 94.1 percent accuracy, 90.6 percent accuracy for two violent individuals and 84 percent for five.
Plus, the paper doesn't mention the false positive rate at all. How often might a drone mistake people horsing around, or playing sports, for something more violent?
Still, this research can be seen as promising work in crowd control -- or proof that we're inching closer to a dystopian future.