A new method of shot change detection based on motion inertia

To resolve video browsing and indexing based-on content, both shot change detection and camera motion recognition are important. From the motion perspective, this paper proposed an approach to extract the frame features after estimating camera motion parameters of frames, combining with the inertial idea from compressed stream. Shot change detection can be achieved by checking the variation of these frame features. Moreover, meaningful and obvious camera motion trends can be recognized during many consecutive frames. The method proposed in this paper is tested on some kinds of standard MPEG stream. The result shows that it can detect not only cut changes, but also gradual transitions. But it´s a pity that more other information is needed when classifying the type of gradual transition. In the aspect of cut change detection, the recall rate is above 90%, and the precision rate is higher than 80%. These are corresponsive to a good performance.