A motion-location based indexing method for retrieving MPEG videos

Abstract

Many content based video data indexing and retrieval systems have utilized the syntactic contents of the video sequences such as color, texture, and shape. And only a few research projects have addressed indexing and retrieving video data by motion. We present a method to automatically compute motion based indexes for video units and to retrieve video units using motion information. Since video is full of motion and the field of motion is the whole frame, we propose a method to divide this field of motion into several subfields and extract the motion features of every subfield. As a result, location information will be added to the motion vectors. Three keyframes will be extracted as representative frames for the corresponding shot, first, last, and middle frame. Then, the system extracts the MPEG coded motion vector information of the macroblocks of the representative frames and uses this information to compute the motion indexes of the corresponding shots. This method is faster than other methods since the motion vectors have already been computed by the MPEG software. Motions that result from camera movement and object movements are distinguished. As a result, users are able to issue queries based on either type of motion.