Robust Video Stabilization based on Motion Decomposition

Video stabilization aims to eliminate camera jitter and improve the visual experience of shaky videos. methods often ignore active movement foreground objects camera, may result in distortion over-smoothing problems. To resolve these issues, this paper proposes a novel video method based on motion decomposition. Since inter-frame is different from that background, we separate feature points background by modifying classic density spatial clustering applications with noise (DBSCAN). The consistent which can be decomposed into camera. And objects. Based decomposition, design first-order second-order trajectory smoothing constraints high-frequency low-frequency components jitter. reduce content distortion, shape-preserving constraints, regularization are taken generate stabilized views all points. Experimental results demonstrate effectiveness robustness proposed variety challenging