Subspace video stabilization based on matrix transformation and Bezier curve

Abstract

Video stabilization improves video quality by removing undesirable jitter to receive stable and comfortable video sequences. This paper proposes a new approach for subspace video stabilization. To make feature trajectories factorization more accurate, we segment feature trajectories into fragments to construct local trajectory matrices, then obtain smooth trajectories based on subspace constraint, matrix transformation and Bezier curve. Finally, according to the original feature points and the final corresponding stable feature points, we use mesh warp to receive high-quality and plausible videos. Experiments show that our method can generate comparable results with regard to some other state-of-the-art video stabilization methods, furthermore in some scenes our results are better than theirs.