Motion segmentation with missing data using PowerFactorization and GPCA

摘要: We consider the problem of segmenting multiple rigid motions from point correspondences in affine views. cast this as a subspace clustering which motion each object lives dimension two, three or four. Unlike previous work, we do not restrict subspaces to be four-dimensional linearly independent. Instead, our approach deals gracefully with all spectrum possible motions: two-dimensional and partially dependent fully In addition, method handles case missing data, meaning that tracks have visible images. Our involves projecting trajectories points into 5-dimensional space, using PowerFactorization fill data. Then linear representing independent is fitted R5 GPCA. test algorithm on various real sequences degenerate nondegenerate motions, perspective effects, transparent etc. achieves misclassification error less than 5% for up 30% data points.