Common motion map based on codebooks

Abstract

This article presents a method to learn common motion patterns in video scenes, in order to detect abnormal behaviors or rare events based on global motion. The motion orientations are observed and learned, providing a common motion map. As in the background modeling technique using codebooks [1], we store motion information in a motion map. The motion map is then projected on various angles, allowing an easy visualization of common motion patterns. The motion map is also used to detect abnormal or rare events. © 2009 Springer-Verlag.