Spatiotemporal Models of Human Activity for Robotic Patrolling

We present a method that allows autonomous systems to detect anomalous events in human-populated environments through understating of their structure and how they change over time. We represent the environment by temporary warped space-hypertime continuous models derived from patterns of changes driven by human activities within the observed space. The ability of the method to detect anomalies is evaluated on real-world datasets gathered by robots over the course of several weeks. An earlier version of this approach was already applied to robots that patrolled offices of a global security company (G4S).