A New Metric between Superquadric Models for 3-D Object Recognition

• Main Authors: Liu,Yao-Tsorng, 劉曜悰
• Other Authors: Lin,Ferng-Ching;Liao,Hong-Yuan
• Format: Others
• Language: zh-TW
• Published: 1993
• Online Access: http://ndltd.ncl.edu.tw/handle/67069398175280019439

碩士 === 國立臺灣大學 === 資訊工程研究所 === 81 === Superquadric models with parametric deformations (including tapering, bending, and cavity deformation, etc.) are suitable models to be used as solid primitives for describing a complicated 3-D shape. This representation scheme is widely used in applications of computer vision and computer graphics during the past decade. Some different methods for recovery of superquadric primitives from range images have already been proposed. But effective distance measures between two superquadric models for the matching task in 3-D object recognition systems are still deficient. All available metrics are error-of-fit measures between a set of range points and a superquadric surface, used by recovery procedures. In this thesis, we propose a distance measure to evaluate the degree of shape similarity between two arbitrary 3-D object models. This distance measure is proved to be a metric. The value of this metric is based on the computation of the minimal total volume of regions bounded between two 3-D object's surfaces. With nice mathematical properties, superquadric models are very suitable for the computation of this metric. Experimental results demonstrate that this metric is effective for matching a recovered superquadric with a database of superquadric models.