基于二维特征轨迹平滑的视频稳定算法

Video Stabilization Algorithm Based on Two-dimensional Feature Trajectory Smoothing

抖动视频不仅影响观看体验,还会加大对后续视频分析编码的难度,因此对抖动视频进行稳像具有重要意义。提出了一种对子图像相机运动轨迹进行平滑的优化框架,有效地稳定了视频序列,避免了过度裁剪,以及最小化各种几何变形。首先对子图像进行特征点检测;运用金字塔光流法进行特征跟踪;采用改进的RANSAC算法剔除特征点误匹配对;然后通过特征点获得的单应性模型来拟合相机轨迹。再对相机轨迹采用结合了双边滤波的连续性保持思想来自适应控制平滑强度的框架进行处理。实验结果表明,提出的算法具有较好的视频稳定效果,同时避免了过度裁剪。

Jitter video not only affects the viewing experience,but also increases the difficulty of subsequent video analysis and coding.Therefore,image stabilization of jitter video is of great significance.In this paper,an optimization framework for smoothing the motion trajectories of sub-image cameras is proposed,which effectively stabilizes the video sequence,avoids over-clipping and minimizes various geometric deformations.Firstly,feature points are detected in sub-images;feature tracking is carried out by pyramid optical flow method;error matching pairs are eliminated by improved RANSAC algorithm;and then the camera trajectory is fitted by homography model obtained by feature points.Secondly,the camera trajectory is processed in a framework that combines the idea of continuity preservation of bilateral filtering to adaptively control the smoothness intensity.The experimental results indicate that the proposed algorithm has good video stabilization effect and avoids over-clipping.