The Bordiga surface as critical locus for 3-view reconstructions (Q1994122): Difference between revisions

In Computer Vision, from a particular scene consisting of images of projected objects, the reconstruction problem asks to recover the position of such objects as well as the projections. In this process, we can find in the projective space of the objects, sets of points for which the projective reconstruction fails; theses configurations of points constitute varieties called the critical loci. In this paper, the authors work with projections both of points from \(\mathbb{P}^4\) to \(\mathbb{P}^2\) and of lines from \(\mathbb{P}^3\) to \(\mathbb{P}^2\). Since three views is the minimum number needed to reconstruct a scene, they consider three projections. In the case of projections of points, the approach used here is to compute the equations of the critical locus by means of the Grassmann tensors introduced in [\textit{R. I. Hartley} and \textit{F. Schaffalitzky}, Lect. Notes Comput. Sci. 3021, 363--375 (2004; Zbl 1098.68775)]. The authors prove that, in the general case, the ideal of this locus defines either a Bordiga surface (the image of the embedding in \(\mathbb{P}^4\) of the blow-up of \(\mathbb{P}^2\) at \(10\) general points by means of the complete linear system of the quartics through these \(10\) points) or a scheme in the same irreducible component of the associated Hilbert scheme. They also prove that every Bordiga surface is the critical locus for the reconstruction for suitable projections. The Bordiga surface is also related to another classical problem in Computer Vision that is the reconstruction problem of a scene in \(\mathbb{P}^3\) consisting of a set of lines (a subset of the Grassmannian \(\mathbb{G}(1, 3)\)). In the case of projections of lines, the authors, as they say in the abstract, compute the ideal that defines the critical locus. This is ``the union of \(3\) \(\alpha\)-planes and a line congruence of bi-degree \((3, 6)\) and sectional genus \(5\) in the Grassmannian \(\mathbb{G}(1, 3) \subset \mathbb{P}^5\)'', moreover, it is biregular to a Bordiga surface [\textit{A. Verra}, Manuscr. Math. 62, No. 4, 417--435 (1988; Zbl 0673.14026)]''. This fact is used to give a bridge between the two reconstruction problems and an algorithm to compute the projection matrices is also described.