Superquadric-based reconstruction from 2D images

Abstract

An approach for superquadric-based reconstruction from 2D images is developed, which computes and reconstructs the parametric models of the VE objects from the 2D images of the counterpart real objects. The superquadric-based hierarchical description as the universal parametric description of the 3D objects is proposed. On one hand, according to the complexity of the real objects, the superquadric curves and their deformations are implemented to fit the contour data of projection images, and the superquadric-based shape descriptions of the real objects are obtained according to the spherical product of superquadrics. On the other hand, the actual size and position are calculated by using epipolar technique for vision reconstruction from multiple uncalibrated images. Combining the two sets of parametric data, the whole superquadric-based descriptions of the real objects are obtained.