Virtual Motion Estimation and Video Stabilization Based on Feature Trajectories
Video stabilization is one of the most important problems of video processing, aimed at removing annoying perturbations in the video so that the output appears more stable. The most challenging videos are those with complex scenes, severe shake, or moving objects. How to effectively and efficiently stabilize above video sequences remains a difficult task. In this paper, we propose an efficient solution based on feature trajectories. Two key ideas are time scalability for trajectory extension and robust estimation for trajectory selection. First, SURF trajectories are used to generate a virtual global motion for stabilization, and then time scalability is introduced for more robust trajectories. Moreover, we alleviate the impact of moving objects through robust estimation. Experiments show our method can reach similar results as the Deshaker. Our algorithm is quite efficient compared with some other algorithms.
Video stabilization motion estimation feature trajectories robust estimation multiple time scales
Qili Chen Li Song Songyu Yu XiaoKang Yang
Department of Electronic Engineering,Shanghai Jiao Tong University Institute of Image Communication and Information Processing Shanghai,China
国际会议
太原
英文
643-647
2011-02-26（万方平台首次上网日期，不代表论文的发表时间）