Representation and Reconstruction of Three-Dimensional Objects Using Nonlinear Deformable Superquadric Models

We propose a new kind of models, which are called Nonlinear Deformable Superquadric (NDS) models, for representing and reconstructing 3D objects. Instead of linear tapering deformation, a piecewise cubic Bezier spline is used to deform superquadrics. By changing the control points of the spline, the shape of the model is deformed as if it was made of a flexible material. In this manner, we can uniformly handle a large set of objects or parts with a small set of parameters. We developed an algorithm for NDS model reconstruction from single monocular image. We first capture the global shape features of the object by extracting the size of a superquadrics. And then an adaptive algorithm automatically fits a piecewise Bezier curve to the image contour. Using the spline to deform the superquadrics, a 3D object with irregular, local shape features is reconstructed. Experimental results using real images show that the reconstructed models are well conformed to the data and the reconstruction procedure is fast.