基于视角优化的便携像机视频去抖算法

Orientation Optimization Based 3D Video Stabilization for Portable Cameras

摘要: 针对便携式摄像机的视频去抖问题，本文提出一种基于摄像机视角优化的视频去抖算法。首先，通过struc- ture-from-motion算法估计摄像机连续运动位姿。其次，基于摄像机的空间位置进行多项式曲线拟合，得到摄像机运动的虚拟路径。然后，通过向量插值平滑视角朝向。最后，基于相机运动平滑曲线与平滑视角对视频帧进行筛选和修正实现视频去抖。实验证明，该算法很好地解决了2D去抖算法对摄像机运动参数估计不足的问题。在真实数据集上的实验结果验证了本文算法在针对便携式摄像机视频数据的去抖有效性。

Abstract: We present a novel view optimization algorithm in the paper for video stabilization. Most existing 2D video stabilization methods may fail to estimate the motion parameters of cameras. In order to solve above problems, we propose a video stabilization method based on 3D technology. Firstly, recovers the original 3D camera motion using the structure-from-motion system. Then, a virtual camera path is computed by polynomial curve fitting. And then, smoothes the view orientation by vectorial interpolation. Finally, corrects the frames which selected in conformity with the smooth- ing location and orientation of cameras. Our experiments on stabilizing challenging videos of real scenes demonstrate the effectiveness of our technique.