3D Shape Analysis

6Rigid Registration

6.1 Introduction

Rigid surface registration is a fundamental problem in computer vision. The goal is to find a rigid transformation (i.e. a combination of translations, rotations, and scaling) which aligns two or multiple 3D surfaces to a common reference frame. Rigid surface registration works as a crucial intermediate step for various vision systems. For example, in a 3D modeling system, point clouds of an object acquired from different viewpoints are transformed into a common reference frame to produce a complete 3D model. In 3D object recognition, a candidate 3D model is registered to a scene with occlusion and clutter to determine if an object is present in the scene. Registration has numerous applications in various areas including robotics, reverse engineering, mould fabrication, industrial inspection, artefact reproduction, virtual museums, games, remote sensing, and medical imaging 155–157.

Existing surface registration methods can broadly be divided into two categories: (i) coarse registration and (2) fine registration 155. In general, coarse registration is first applied to estimate a rough initial transformation between two surfaces without any prior alignment. The estimated transformation is then refined by fine registration to produce a more accurate alignment between two surfaces 87. For coarse registration, either manual or automatic methods can be used. A manual method usually uses a turntable, robot arms, positioning devices, or attached markers ...