TL;DR本文介绍了一种结合多种模型的多视角谱聚类框架,旨在解决 fundamental matrix 在运动分割中的缺陷,通过测试运动分割数据集,我们证明了这一方法在实际应用中的有效性。
Abstract
Many real-world sequences cannot be conveniently categorized as general or
degenerate; in such cases, imposing a false dichotomy in using the fundamental
matrix or homography model for motion segmentation would l