Automatic video segmentation in the compressed domain

In the current military operational environment, an increasingly large quantity of motion imagery data is being collected by both manned and unmanned airborne surveillance platforms. The Department of Defense has made a commitment to move the collection of this data into the digital domain to support more effective and efficient exploitation, targeting MPEG-2 as the decompression algorithm of choice. A major impediment to the exploitation and management of this motion imagery is the sheer volume of that data. With more information continually being provided to fewer people for exploitation, techniques need to be developed that will provide the analyst with manageable amounts of coherent video. Automatically segmenting the video into manageable duration clips is one way to achieve this. This paper describes an automatic, scene based, segmentation algorithm.