Video analytics is software that watches security camera video by itself and raises an alert when it sees a problem, such as a person following someone else through a door, a person in a closed area at night, or a weapon. It is also called AI video analytics or video content analysis. It uses computer vision, the field of AI that recognises people, vehicles and objects in images.
In a large building, the security team cannot watch hundreds of screens at once. Studies of guards show that attention drops within minutes of watching screens. So most video is never seen by anyone until something has already gone wrong. Video analytics changes this. It runs over the cameras that a site already has, checks every frame, and passes on only the few moments that need a person.
Consider the door to the server hall of a bank's data centre in Navi Mumbai. The rule is that each person taps a card. At 11:20 p.m., an engineer taps his card and the door opens once. But the camera above the door shows two people walking through. The second person did not tap. This is called tailgating. The analytics software compares the number of people it counted, two, with the number of card taps reported by physical access control, one. It sends an alert with a ten-second clip to the duty guard's phone. The guard watches the clip, sees that the second person is a contractor without a card, and walks to the door.
The models make mistakes. A shadow, a reflection or a trolley can look like a person. Each alert therefore goes to a person, who checks it before anyone acts. Sorting real alerts from false ones is called alert triage. Serious alerts, such as a weapon at the gate, go to critical event management, which warns everyone nearby.
The analysis can run in two places. It can run in the cloud, on the video sent there by the video management system (VMS). Or it can run on a small computer near the cameras, which sends only the alerts. This second way is called edge AI, and it needs much less internet bandwidth.
Indian cities also use video analytics on roads. Cameras at junctions read the number plates of vehicles that jump a red light, and the city's traffic police send an e-challan, a fine sent by message, to the owner. Reading a plate from video is called automatic number plate recognition (ANPR). Face recognition in public places is more disputed, because it identifies people who have done nothing wrong.