An object-centered three-dimensional model builder

Abstract

A method of building a three-dimensional model of a rigid object using information from many views is described. Planar and quadric surface primitives describe the object's surface in an object-centered reference frame. The extent of a primitive is defined by the intersection of the primitive with its neighbors. An edge graph defined by these intersections implicitly expresses spatial relationships between surface primitives. The model builder's input consists of groups of data points corresponding to different views. Each data point contains spatial and orientation information about the object's surface at a discrete location. A set of registered arrays is used to summarize input information in local areas. Mathematical principles from differential geometry are applied to determine local surface properties. A region-growing technique is applied to this information to identify data points which then are represented by a surface primitive. Edges and corners are computed based on the intersections of surface primitives. The results from the analysis of the various views are transformed to a common, arbitrary reference frame for integration into a global model. The final object-centered reference frame is established based on the center of gravity and moments of inertia of the object as determined from the complete model. The goal of model building has applications in the fields of pattern recognition, computer vision, robotics, computer-aided design and computer-aided manufacturing. A model using surface primitives appears as a natural first step in describing an object because surfaces are obvious visual features. The strengths and weaknesses of this surface model are explored.