一种基于均值偏移的自动运动分割方法

Automatic Motion Segmentation of Sparse Feature Points with Mean Shift

运动分割是计算机视觉领域研究的重要内容。提出一种基于均值偏移的自动运动分割算法。该方法首先用特征点匹配关系获得特征点的运动轨迹,并以轨迹的运动向量作为特征,再用均值偏移算法对轨迹的运动向量进行聚类。均值偏移缩小相似的运动向量之间的差别,同时扩大不同运动的运动向量之间的差距。为了自动获得运动分类数,还提出了一种基于非参数核密度的自动分类方法,该方法通过估计运动向量的密度分布,用核密度图自动确定运动分类数。实验结果表明,该算法分割精度高、鲁棒性好,能够自动确定运动分类数。

We proposed an automatic motion segmentation operating on sparse feature points. Feature points are detec- ted and tracked throughout an image sequence, and feature points are grouped using a mean shift algorithm. The motion segmentation is driven by the density of the motion vector in feature space. The kernel density estimation is performed on the mean-shifted motion vector and the number of motion present is estimated by the number of peaks in the kernel density curve. Experimental results on a number of challenging image sequences demonstrate the effectiveness and ro- bustness of the technique.