Video stabilization by sparse and low-rank matrix decomposition

Bo Gao,Ziming Kou,Zemin Jing
DOI: https://doi.org/10.1109/CAMAN.2011.5778768
2011-01-01
Abstract:This paper presents a novel approach to stabilize video sequences based on low-rank matrix decomposition. Compared to previous methods which are based on simplified models, our stabilization system can work in situations where significant depth variations exist in the scenes and the camera undergoes large translational movement. We formulate the stabilized frames as a low-rank matrix. This allows us to precisely control the smoothness by decomposing low-rank matrix. By taking advantage of the sparseness, our optimization process is very efficient. Instead of recovering dense depths, we use approximate geometry representation and analyze the resulting warping errors. We show that by appropriately constraint warping error, visually plausible results can be achieved even using planar structures. A variety of experiments have been implemented, which demonstrates the robustness and efficiency of our approach. ©2011 IEEE.
What problem does this paper attempt to address?