Background

When I’m at the library, I’d like to be able to go to the toilet, without collecting all my stuff from the table. Part of the solution is to have a camera installed that films all the tables, but assuming we can hire someone to look at the camera-feeds, that person might not notice that my laptop was stolen. Of course they could be notified, and the culprit identified from the tapes, but what if the culprit is “disguised? The only solution is to capture the thief before he/she leaves the building. For that to work, the security personel must be notified of the theft exactly when it happens!

Formal problem definition

Problem:

• Person A, me, leaves an artifact (computer) at a table in a public space and goes somewhere (restroom).
• Person B drops by table and steals computer
• By the time A is back, B has left the building
• Because B was wearing sunglasses and a blue beard, B can not be identified from the surveillance tape
• Person A is sad

Solution requirement: Person B should be apprehended before he/she leaves the building, namely before person A is back and notices the theft. This means that an algorithm must detect the theft as it happens!

Solution approach:

• Camera feed is routed to bank of algorithms
• Algorithm X detects people and their location, and assigns unique IDs to different people
• Algorithm Y detects artifacts, and associates each artifact with the ID of its owner
• Algorithm Z detects the situations: 1) An owner has lefts his/her artifact 2) A person which is not the owner is very near an artifact. If both 1 and 2 hold for a given artifact, an event is fired

The events from algorithm Z are handed to the security staff, who can investigate visually whether a theft is taking place.