2024

KATHLÉN KOHN

Department of Mathematics, Royal Institute of Technology
A person's portrait
project title

Algebraic Vision

funded by

Ragnar Söderbergs stiftelse

Hon utforskar djupet i 3D-rekonstruktion genom matematik

 

Idag dominerar kameror med rullande slutare (RS) marknaden tack vare sitt överkomliga pris, förbättrad upplösning och snabbare bildhastighet. Dessa kameror styr också teknologin inom områden som assisterad körning, med sin förmåga att exponera sensorer rad för rad över tiden, istället för en hel bild i en enda sekvens.

 

Traditionella rekonstruktionsmetoder står dock handfallna när kameran är i rörelse, särskilt när den är monterade på snabbt rörliga plattformar som bilar, vilket är vanligt förekommande idag.

 

Kathléns projekt utforskar den komplexa världen av 3D-rekonstruktion från bilder fångade av kameror med rullande slutare. Genom att fördjupa sig i algebraisk geometri, skapar Kathlén en matematisk grund som är skräddarsydd för just dessa kameror. Hennes arbete banar väg för snabbare rekonstruktionsalgoritmer som överträffar noggrannheten hos metoder som förlitar sig på maskininlärning.

 

 

Algebraic vision is the two-way street between algebraic geometry and computer vision.

The primary goal of this project is to develop the theoretical foundations for 3D scene reconstruction from images taken by unknown rollingshutter cameras, which is the overwhelming camera technology of today. Implementing sufficiently fast reconstruction algorithms for rolling-shutter cameras – without restricting assumptions – is a major open challenge in computer vision.

 

Algebraic geometry provides the natural tools for rigorous theoretical foundations for that challenge, yet the algebraic community has not investigated rolling-shutter cameras.

From the algebro-geometric perspective, rolling-shutter cameras are parametrizations of algebraic surfaces in the Grassmannian of 3D lines, and 3D reconstruction amounts to computing fibers under rational maps. AVIS will use this inherent geometry to find exhaustive lists of efficiently solvable algebraic reconstruction problems (so-called minimal problems) and develop new intersection-theoretic tools to measure their intrinsic complexity.

 

This project (AVIS) will describe the critical loci of reconstruction where problem instances are ill-conditioned and prone to numerical instability. Moreover, AVIS will compute the polynomial constraints encoding relative camera poses, and calculate geometric invariants that capture the complexity of triangulation (i.e., the problem of recovering 3D coordinates from known cameras and 2D data) under noise.

 

If successful, the algebro-geometric foundations developed in this project will lead to the implementation of fast 3D reconstruction algorithms with rolling-shutter cameras. AVIS’ unified and abstract view will yield tools applicable to 3D reconstruction in general, beyond rolling-shutter cameras, and lead to long-lasting foundational results in modern computer vision. The mutual exchange between algebraic geometry and computer vision will open new horizons for scholarship in both fields.

Foto: Emma Burendahl

April 2024

Meet the SFSG fellows

See all