CAD-Based Object Reconstruction Using Line Photogrammetry for Direct Interaction between GEMS and a Vision System

The paper presents an algorithm for reconstructing Geometric Element Modeling System ( ~ E ~ ~ l b a s e d industrial objects using line photogrammetry (LP) in order that there can be direct interaction between GEMS and a vision system. In the GEMS system, each object is constructed by Boolean set operators of constant solid geometry (CSG) primitives, each face of which is described by a Boundary presentation (B-pre); and straight lines, curves, planes, and curved surfaces are described by parametric equations. Based on this type of data structure and representation, the geometric elements which describe industrial objects in GEMS, are taken as unknowns, and are directly solved for by matching images (20) and objects (30). Compared to traditional photogrammetric DEM products, the advantage of this algorithm is that the product of reconstruction is parameters describing the industrial object, which can be directly fed back to the GEMS system. It is convenient to communicate between GEMS and the vision system without any transformation. On the other hand, the developed mathematical model can be used not only for lineto-line correspondences but also for line-to-surface or surfaceto-surface correspondences. A number of simulations and practical experiments demonstrated that this algorithm is feasible and reliable in industrial computer vision. Introduction Our vision research group at the Department of Computer Science and Technology of Tsinghua University was designing a vision system for the Chinese Aerospace Industrial Department. The objective of the system was to provide sufficient and reliable vision information for an Experimental Testbed of Space Intelligent Robotic Manipulators. We call this system the Space Intelligent Vision Equipment (Srm) (Yu, 1995). One of tasks in SIVE is to reconstruct Geometric Element Modeling System (GEMS)-based in industrial objects in order that the operator of SNE can manipulate and locate a robot arm by comparing the reconstructed object and one represented in GEMS (GEMS, a type of CAD system, is described in detail in the next section). However, the traditional product of reconstruction is a DEM or DSM (digital surface model); and this form of product cannot be fed back into the GEMS system directly. On G. Zhou was with the Department of Civil & Environmental Engineering and Geodetic Science, The Ohio State University, 470 Hitchcock Hall, 2070 Neil Avenue, Columbus OH 432101275. He is presently with the Department of Civil Engineering and Technology, Kaufman Hall, Room 214, Old Dominion University, Norfolk, VA 23529 ([email protected]). D. Li is with the Department of Photogrammetry and Remote Sensing, Wuhan Technical University of Surveying and Mapping, Wuhan 430079, China. the other hand, when reconstructing these industrial objects, the following problems are encountered: 11) except for geometric features (e.g., lines and edges), distinct features can seldom be found on a surface; and (2) it is difficult to find 2D-to-ZD corresponding points due to the vacuum of space, non-uniform lighting, object drift or instability, and the dynamic SIVE environment in space. Obviously, the traditional approach of pointto-point correspondence is not suitable for this real situation. Thus, we had to develop an algorithm which could reconstruct 3D industrial objects based on the prior knowledge of GEMS when considering the above real situations. For this reason, this paper presents an algorithm for reconstructing GEMS-based industrial objects using line photogrammetry (LP). This is because 3D spatial lines (straight or curved) and surfaces (planar or curved faces) are valuable and available because their ZD projections in images are more easily extracted than are point features in industrial images; line features (the term line features refers to one-dimensional straight lines and curves, or to the projection of spatial lines or spatial surfaces) are more robust, are much more geometric, and contain more semantic information than do point features; an object is constructed by combining Construct Solid Geometry (CSG) and B-presentation (El-pre) in GEMS. Each primitive is described by parameters; and the data structure and topologic information for industrial obiects in GEMS are available. Based on the above characteristics, we first developed a mathematical model for reconstructing a single primitive, and then went on to develop a mathematical model for reconstructing complex objects while considering the data structure and topological information contained in GEMS. In the model for reconstructing a single primitive, geometric elements, which represent industrial objects in GEMS, are taken as unknowns, and are directly solved for by matching images ( 2 ~ ) and objects ( 3 ~ ) . The reconstruction of complex objects will make full use of prior knowledge of the data structure and representation in GEMS as well as the mathematical models for reconstructing an individual primitives. Some references in the photogrammetric community to previous research and practical work based on line-to-line correspondences in computer vision and photogrammetry include Doehler (1975), who used horizontal and vertical lines as control features for close-range photogrammetric Photogrammetric Engineering & Remote Sensing Vol. 67, No. 1, January 2001, pp. 107-116. 0099-1112/01/6701-107$3.00/0 O 2001 American Society for Photogrammetry