DBSCAN data clustering algorithm for video stabilizing system

This paper proposed a method based on DBSCAN data clustering algorithm to stabilize the jitter of digital video with moving objects in it. In order to distinguish the corners on moving objects with those on background, after extracting the corners of each frame, DBSCAN algorithm was used to cluster all the corners by clustering their motion vectors´ lengths and directions. Then we compared the dispersion of each cluster to confirm whether the corners in each cluster were belonged to moving objects or background. Simulation experimental results showed that the proposed method had good stabilization effects to stabilize jitter in video sequence with moving objects in it.