Chapter 61 — Robot Surveillance and Security

Tracking people for security

Tracking of people in crowded scenes is challenging because people occlude each other as they walk around. The latest revision of the University of Minnesota's person tracker uses adaptive appearance models that explicitly account for the probability that a person may be partially occluded. All potentially occluding targets are tracked jointly, and the most likely visibility order is estimated (so we know the probability that person A is occluding person B). Target-size adaptation is performed using calibration information about the camera, and the reported target positions are made in real-world coordinates.
Nikos Papanikolopoulos
Thanks to the Artifical Intelligence, Robotics and Vision Laboratory at the University of Minnesota