摘要
针对基于传统高斯滤波的数字稳像算法在运动平滑时使用固定权重,限制了对视频抖动的滤除能力问题,提出了基于无累积误差滤波的数字稳像算法。在运动估计中,对视频帧的特征点进行提取和匹配,基于仿射模型利用匹配的特征点估计两帧间的全局运动;在运动平滑中,直接对当前帧与其前后固定半径范围内相邻帧的运动估计参数进行均值滤波,避免了累计误差的问题。通过滤波在滤除抖动的同时,得到平滑的运动。实验证明,该方法具有更好的视频抖动滤除能力及良好的稳像效果。
To solve the inefficient of Gaussian filtering that perform motion smoothing by a constant Gaussian kernel,this paper proposed a novel low-pass filter based digital video stabilization method without cumulative error. In motion estimation step,it extracted and matched the features,and estimated the affine motion transformation parameters between two frames using matched features. In motion smoothing step,instead of reference frame,it used mean filtering locally to smooth the transformation parameters from the current frame to the neighboring frames,which was free from cumulative error. It removed jitter and in the same time also obtained the smoothed motion by a low-pass filter. The experiments show that the proposed method has the superior motion smoothing ability and good stabilization performance.
出处
《计算机应用研究》
CSCD
北大核心
2014年第7期2213-2215,2223,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(61071162)
关键词
数字稳像
运动平滑
累计误差
均值滤波
digital video stabilization
motion smoothing
cumulative error
mean filter
