Motion segmentation using feature selection and subspace method based on shape space

Abstract

Motion segmentation using feature correspondences can be regarded as a combinatorial problem. A mo tion segmentation algorithm using feature selection and subspace method is proposed to solve the combzna to rial problem. Feature selection is carried out as com putation of a basis of the linear space that represents the shape of objects. Features can be selected from "each" object "without segmentation information" by keeping the correspondence of basis vectors to features. Only four or less features of each object are used; the combination in segmentation is reduced by feature se lection. Thus the combinatorial problem can be solved without optimization. The remaining features in se lection are classified using the subspace method based on the segmentation result of selected features. Ex periments are done to consider the usefulness of the proposed method.

Improved Multistage Learning for Multibody Motion Segmentation

“…The lower row shows our 3-D visualization of the trajectories. Table 1 lists the correct classification ratios at each stage of our method 4 and some others: the MSL of Sugaya and Kanatani 5 [18]; the method of Vidal et al 6 [22]; RANSAC 5 ; the method of Yan and Pollefeys 5 [25]. We can see that for all the videos, our method reach high classification ratios in relatively early stages and 100% in the end, while other methods do not necessarily achieve 100%.…”

“…Ichimura [5] used the Otsu discrimination criterion. He also used the QR decomposition [6]. Inoue and Urahama introduced fuzzy clustering.…”

“…http://www.vision.jhu.edu/data/hopkins155 4 http://www.iim.ics.tut.ac.jp/˜sugaya/public-e.html5 The code is at the cite in the footnote 4 6. We used the code placed at the cite in footnote 3.…”

Geometric Structure of Degeneracy for Multi-body Motion Segmentation

“…Eq. (5) implies that the trajectories of the feature points that belong to one object are constrained to be in the 4-D subspace spanned by {m 0 , m 1 , m 2 , m 3 } in R 2M . It follows that multiple moving objects can be segmented into individual motions by separating the trajectories vectors {p α } into distinct 4-D subspaces.…”

Removing Mistracking of Multibody Motion Video Database Hopkins155

“…Ichimura [7] used the discrimination criterion of Otsu [16]. He also used the QR decomposition [8]. Inoue and Urahama [9] introduced fuzzy clustering.…”