基于3D旋转模型的全景视频稳像算法

A PANORAMA VIDEO STABILIZATION ALGORITHM BASED ON 3D ROTATION SMOOTHING

视频稳像算法是指消除由于手持相机拍摄造成的视频帧间抖动的算法。由于全景视频广视域的特性,传统的稳像算法并不适用。针对全景视频,提出一种使用3D旋转模型进行稳像处理的新方法。通过特征点提取跟踪计算出的3D旋转模型可以准确地描述相机的真实运动路径;为了减少数据冗余,使用欧拉角表征旋转模型的参数。在路径平滑过程中,引入一种基于样条拟合的改进方法,分离抖动和主观运动并消除高频的抖动,相比传统方法更加准确可靠。实验结果表明,该算法可以有效地消除视频中的抖动并生成稳定的视频序列。

Video stabilization seeks to remove unwanted frame-to-frame jitter due to camera shake.Due to the wide field of view of panorama video,the traditional video stabilization algorithm is not suitable.In this paper,we propose a novel approach for stabilizing panorama video by a pure 3D rotation motion model.The 3D rotation which estimated by feature tracking,can reflect real camera motion accurately.We described the rotation parameters by Euler angles instead of projection matrix,for reducing data redundancy and computational complexity.Moreover,we introduced an improved method based on cubic spline to separate jitter and subjective motion and eliminate high frequency jitter,which was more accurate and reliable compared with traditional methods.The experimental results on a variety of real video sequences show that our approach can remove annoying jitter effectively and generate stable video.