- Activity detection. If something interesting is happening in front of a camera (e.g. two people running in front of a camera middle of the night), AI detects that and pops the view up in front of humans for further action. Because less than 1% of cameras contain any useful info at any moment, they don't need that many humans to watch those cameras live.
- Face recognition. For many cameras, Face capture and recognition is running live and report any criminals or targeted personnel to police.
- Footage markup. If police need to go through the camera footage manually, the recording playback can skip the uninteresting parts to save time.
Combining these with unrestricted real-time integration of other data sources, such as real time GPS from mobile apps, cell tower call history from telecoms, network traffic inspection and remote spying capabilities, Chinese police forces can pretty much find anybody very quickly. They are also known to have state of the art big data platforms developed in house.
One anecdote I read: Somebody killed a person in a small city middle of the night, removed battery of his phone, ran to his car parked on street, drove a couple of hundred miles to another middle sized city, only to be caught in a motel next morning. How? Complete camera footage covering his walking path, vehicle plate tracking all the way to his destination, and motel check-in system that is also integrated with police.
Will some AI monitor all the feeds and flag suspicious activities to the local authorities?