レンジ情報を用いた内部放射光の投影による形状表現の研究

抄録

The possibility to utilize 3D information for engineering purposes is incleasing by the recent development of various 3D sensors. But because optical 3D sensors such as laser range finder use light, which moves straight, the measurement area is limited to the front face of the object and the back side is not measurable. To adapt such unmeasurable area, we need a system that memorizes shapes of objects that have often encountered and superimposes them to scene. To realize such kind of system, an appropriate 3D shape representation, which is 1) able to compare partial and entire shape information and to detect corresponding areas 2) working as quick as possible for real-time tasks, is needed. We propose a novel representation framework to describe and compare 3D objects: Internal Radiatedlight Projection (IRP) - projection of local shape informations of an object on a sphere by imaginary rays from the proposing reference point “kernel”-. In order to describe local shape information, Harmonic Contour Analysis (HCA), and to arrange them properly the shape matrix are used respectively in IRP framework. The characteristics of IRP are: (1) simple representation for complicated 3D shape information; (2) use of local shape characteristics and their adjacency informations; and moreover, by using shape matrix, (3) simple matching procedure considering the effect of gravity and stable poses for objects. By IRP representation, we can categorize objects in known classes and estimate their positions and attitudes. In this paper, the basic concept of IRP, a reliable way to obtain reference point “kernel”, representation of local 3D shape by HCA, and shape matrix based comparison and categorization of objects are explained. Experiments of object recognition for both virtual and real objects are shown to demonstrate its efficiency and feasibility.