Providing safety of passengers at the airport is an expensive task as a lot of human operators is required to analyze video streams on line for 24/7. Therefore it seems desirable to automatically monitor passengers and detect those that might present a threat to others or themselves.
Description
The goal is to detect passengers that behave suspiciously at the airport from their trajectories of movement. In one scenario we aim to detect suspicious passengers that go from point A to B while avoiding authorities. We use two-level approach that at the first level detect potentially suspicious actions using HMMs and CHMMs, and at the second level merge several actions into the final decision if behavior of particular passenger is suspicious or not using LHMMs, Utility-Based Plan Recognition and Penalty-based Accomulation. As a result we assign "level of suspiciousness" to each passenger.
Currently, we generate data in ESCAPES, a multi-agent simulator for airport evacuations, and extract 2D trajectories of all passengers.
