摘要
本文以经典的L-K光流方程为出发点,提出了一种高效的基于SVD协方差加权的光流估计算法,并成功应用到柔性目标点跟踪中,有效地解决了传统L-K算法的孔径问题、深度不连续点的估计和长序列视频的漂移问题。基于标准测试序列的试验结果,证明该算法能有效地跟踪较长视频序列中具有2D和1D甚至基本没有纹理的具有退化结构的柔性目标点,同时结果还可以作为半稠密的点对应来解决SFM问题中的一个关键难题correspondence。
A novel algorithm namely covariance weighted SVD based optical flow estimation is proposed in this paper, which achieves more robust and precise tracking result comparing to the traditional L-K tracker. Covariance weighted is used to transform the problem from a hyper-ellipse space to a hyper-sphere space. SVD is correspondingly performed to involve the subspace constraints. The proposed algorithm has been evaluated by the standard test sequence. The potential applications vary from articulated automation, structure from motion, computer surveillance to human-computer interaction.
出处
《计算机科学》
CSCD
北大核心
2006年第6期236-238,共3页
Computer Science
基金
国防基础研究"多源多目标协同感知及超光谱自主检测技术"
航天创新基金"基于图像分析的作战效果分析与评估"的支持。
关键词
视频跟踪
光流估计
协方差加权
SVD
人脸特征点跟踪
Visual tracking, Optical flow estimation, Covariance weighted, SVD, Facial features tracking
