【摘要】 数码相机拍摄的视频往往带有因为拍摄人手的不稳定而引入的抖动,因此该文给出了一个视频抖动矫正方案,包括全局运动估计、运动的抖动分量滤除、抖动补偿3个模块。对摄像机运动模型,该文考虑了最常见的水平、垂直平动和平面转动这3个运动参数来进行全局运动估计,首先采用基于块匹配运动估计方法来得到运动参数的初始值,然后用Newton迭代法进行参数优化;从运动学的角度出发设计了一种以二次B样条方法为基础来拟合摄像机的运动轨迹方案。实验证明这种滤波器的设计方法能够很好地去除视频抖动,明显改善视频的主观效果。该方法不仅简单,而且不需要预知抖动的模式就能够确保运动轨迹的平滑。更多还原

【Abstract】 Video shot with a hand held digital camera may have jitter caused by the holder’s motions. This paper presents an algorithm to stabilize the jitter in videos which includes three modules: global motion estimates, motion filters, and motion compensation. The most common motions, translation, and rotation are currently corrected by the algorithm. Block matching is used to calculate the initial motion parameters which are then refined using the Newton iteration method. A two-order B-spline-based filter is then used to alleviate jitter in the video. Unlike previous techniques, a smooth video can be constructed without a prior jitter model. Test results show that the method removes video jitter and significantly improves the visual effect.更多还原