Distortion varieties (Q667656)

Distortion varieties are motivated by multi-view geometry in Computer Vision. In this field, one can think of a camera as a linear projection from $\mathbb{P}^3$ to $\mathbb{P}^2$ and in a model as a projective variety $X$ in $\mathbb{P}^n$ representing the relative positions of two or more cameras, so that data are the correspondences of image points in the plane. A model for cameras with image distortion allows for an additional parameter and each coordinate of a model is now multiplied by a polynomial in the unknown parameter whose coefficients depending on the data. Fixed a vector $u$ of $n+1$ nonnegative integers and given a projective variety $X$ of dimension $d$ in $\mathbb{P}^n$ (the model), the distortion variety $X_{\left[u\right]}$ of level $u$ and dimension $d+1$ is defined as the closure of the image of the regular map $X \times C \rightarrow \mathbb{P}^N$ given by the parametric representation of the rational normal scroll $S_u$, where $N$ is the sum of $n$ and the $n+1$ nonnegative integers of the vector $u$. Thus, the distortion varieties are parametrized by duplicating coordinates and multiplying them with monomials. In this paper, the authors study the degrees of distortion varieties in terms of the variety and the defining equations. Examples illustrate the results. For the case of one-parameter distortions, exact formulas are obtained and these are based on Chow polytopes and Gröbner bases. For the case of multi-parameter distortions, the role of the scroll is now played by a toric variety and they are studied using tropical geometry. As the authors say in the abstract, their theory furnishes a new framework for formulating and solving minimal problems for camera models with image distortion. They also provide an application of their method to a minimal problem by building and testing an efficient Gröbner basis solver for it. Finally, computational results are presented.