Research Initiation: CSG-Based 3-D Object Recognition UsingRange Images

本研究的目的是开发一个基于构造性立体几何(CSG)的三维物体识别系统的理论和算法。所提出的系统首先将距离图像中的对象分割成原始曲面片。然后，将原始曲面片分类为原始体积的曲面，如球体、圆柱体、立方体或圆锥体。此外，还估计了这些基元的比例、位置和方向。下一步是确定这些原语上的集合操作的顺序和类型。这将使用描述操作顺序的优先图来完成。识别是通过将导出的优先图与数据库中的优先图进行匹配来实现的。这一研究成果将为CAD和视觉数据库的统一提供一种手段，并有助于推进自动化制造系统的技术进步。

The objective of this research is to develop the theory and algorithms for implementation of a constructive solid geometry (CSG) based 3-D object recognition system. The proposed system starts with segmenting an object in a range image into primitive surface patches. Then, the primitive surface patches are classified as surfaces of primitive volumes such as spheres, cylinders, cubes, or cones. In addition, the scales, locations, and orientations of these primitives are estimated. The next step is to determine the order and types of set operations on these primitives. This will be done using a precedence graph describing the order of operations. Recognition is achieved by matching the derived precedence graph with those in the database. Results of this research should provide a means to unify CAD and vision databases and help to advance the technology of automated manufacturing systems.