3-D object representation from multi-view range data applying deformable superquadrics

摘要

This paper presents a new framework for recovering superquadrics with global deformations from multi-view real range data. The framework aims at improving confidence and accuracy of recovered models by utilizing multiview information, and consists of the initial superquadric model recovery, view registration, view integration, and final model recovery from integrated data. A quadrant analysis technique is proposed to aid the recovery of bending superquadrics. A modified range data registration method based on recovered superquadrics is also proposed to handle tapered superquadrics. Experimental results indicate the proposed framework of multi-view representation significantly improved the accuracy and confidence of recovered superquadrics compared with existing recovery strategies which rely on single-view range images.