Non-rigid structure from motion using piece-wise Bézier curves

Andrada Mendoza, Iván

Non-rigid structure from motion using piece-wise Bézier curves

Citation

Andrada Mendoza, I. Non-rigid structure from motion using piece-wise Bézier curves. 2021. handle: http://hdl.handle.net/10230/48621

This citation was generated automatically.

Abstract

Non-rigid structure from motion (NRSfM) is an area of interest that has several years of history within the field of Computer Vision. Such discipline focuses on recovering the 3D geometry of a deforming (therefore, non-rigid) object from its 2D feature correspondence over several frames or images. NRSfM is present at several real-world applications such as animations, data-recovery from camera, medical surgery, among others. Nowadays, there is no general algorithm that can solve this geometry recovery issue for all kinds of conceivable motion. The purpose of this project is mainly concerned in implementing a new way of retrieving the 3D structure from a non-rigid object according to its features extracted from motion using trajectory-based subspaces (also referred as basis). The subspaces used for this aim are a set of piece-wise Bézier curves subspaces constrained with several continuities based on curve smoothness (C0, C1 and C2, regarding continuity and derivability of a Bézier cuve). Such mathematical formulation provides a neat physical interpretation of features, in addition to a smoother and better representation of the motion.