Motion descriptors for content-based video representation

摘要

This paper presents two motion descriptors which were recommended by MPEG to become part of the first visual reference model (XM 1.0) of the evolving MPEG-7 standard in development. These motion descriptors are: (i) the camera motion descriptor which describes the global motion of the camera or of the observer in a natural 3-D scene, and (ii) the object motion trajectory descriptor which describes how an object moves in 3-D space or in the 2-D image plane. These two descriptors are important elements in capturing the dynamic content of video sequences in a compact form. They are used to index video sequences according to their dynamic content. Applications that use these descriptors include TV program classification, video editing for broadcast TV and movies, broadcast sports, and video surveillance.