Fast content-based MPEG video indexing using object motion

We propose a new object motion histogram for content-based indexing of MPEG video data. Video data are segmented into shots. Then object motions are extracted and clustered after removing background motion. We adopted polar coordinates to describe the clustered object motion. An object motion histogram in a shot (OMHS) is defined and obtained on the proposed polar division technique. The first moment of object motion is utilized as a descriptor of object motion.