Critical loci for projective reconstruction from multiple views in higher dimension: a comprehensive theoretical approach (Q486220): Difference between revisions

The problem of image reconstruction has a deep background in Algebraic Geometry. From a theoretical point of view, the initial step consists in recovering sets of points \(Z\subset\mathbb P^3\) from a finite set of projections to \(\mathbb P^2\). When dynamical scenes are considered, then the problem generalizes to the reconstruction of sets \(Z\) in some projective space \(\mathbb P^k\), from planar projections. When the number of projections is low with respect to \(k\), there is no way to reconstruct uniquely the set \(Z\). Even when the number of projection grows, so that the reconstruction becomes possible for general sets, yet there exists a critical variety over which a unique reconstruction of \(Z\) cannot hold. Such varieties are called \textit{critical loci} and their properties depend on the dimension \(k\) of the space and the number \(n\) of projection. The authors' target is the study of critical loci for the reconstruction problem. These loci turn out to be determinantal varieties, whose structure is important for applications to concrete reconstruction algorithms. In particular, when \(n\) is minimal such that the reconstruction holds for general sets \(Z\), then the author provide a description the geometry of critical loci \(X_k\)'s. For \(k=2r-1\) odd, the \(X_k\)'s are special hypersurfaces of degree \(r\), whose structure and degenerations are investigated. When \(k=2r\) is even, the \(X_k\)'s turn out to be determinantal varieties of codimension \(2\).