Industrial geometry: recent advances and applications in CAD
Introduction
During the past decades, geometric methods have played an increasingly important role in a variety of areas dealing with computing for industrial applications; these include Computer-Aided Design and Manufacturing, Geometric Modeling, Computational Geometry, Robotics, Computer Vision, Pattern Recognition and Image Processing, Computer Graphics and Scientific Visualization.
These areas originated from different requirements in specific applications and thus they have seen rather disjunct developments. In fact, very similar problems have been treated by different communities. These communities still have different favorite solutions to nearly the same problems. Let us illustrate this at hand of curve approximation. According to industry standards, the CAD approach uses B-spline curves and a method for data fitting which iterates between parameter estimation and linear least squares approximation [11], [34]. Computer Vision and Image Processing developed another method, active contours [4], [12], which have originally been formulated as parametric curves. Nowadays, the advantages of (discretized) implicit representations and the formulation of the curve evolution via partial differential equations in the level set method [19], [30] are highly appreciated, in particular, for difficult curve approximation problems which arise in image segmentation. Curve approximation also appears in higher dimensional spaces. For example, in the space of rigid body motions, it leads to motion design for Robotics [17] or Computer Animation.
In recent years, these different areas of research have started to become increasingly interconnected, and have even begun to merge. A driving force in this process is the increasing complexity of applications, where one field of research alone would be insufficient to achieve useful results. Novel technologies for acquisition and processing of data lead to new and increasingly challenging problems, whose solutions require the combination of techniques from different branches of applied geometry. The thereby emerging research area, which aims at unifying existing and developing new methods and algorithms for a variety of application areas with a strong geometric component, shall be called Industrial Geometry.
Let us continue the example from curve approximation addressed above. The viewpoint of Industrial Geometry would be to investigate the various algorithms from a common perspective. Since all the available algorithms are solving non-linear geometric optimization problems, it is appropriate to study and compare the known approaches from the optimization perspective. In the present paper, we will point to recent results in this direction.
It is impossible to outline all major current research streams in Industrial Geometry in this paper. Therefore, we will focus just on a few topics. We will briefly look at the level set method [19], [31] and on hybrid data structures for geometric computing [15]. The major part of this paper is devoted to geometric optimization problems which involve distance functions. Here we will present a survey with some new results on a recently developed class of optimization algorithms, which can be called squared distance minimization. The benefits of the optimization viewpoint rather than the perspective of a specific application will become obvious. With nearly the same algorithms we can solve a wide class of curve and surface approximation problems and a number of registration (surface matching) problems.
The methods we are using for the topics indicated so far have a relation to Computer Vision and Image Processing. As a further example for the fruitful use of techniques which originate in these fields, we discuss a new metric on surfaces. It is sensitive to features such as smoothed edges, which are characterized by a significant deviation of the two principal curvatures. This new metric can be easily understood with the concept of an image manifold [14], and it has a number of interesting applications [25]. For example, we can design curves on surfaces whose shape is adapted to the features of the surface. Moreover, we briefly address local neighborhood definition and region growing as an aid in the segmentation process for reverse engineering of geometric objects. Image processing frequently uses mathematical morphology for basic topological and geometric operations [10], [30]; this work describes similar operations on surfaces, which—if desired—can be made sensitive to the features.
Section snippets
Geometry representations
The choice of an appropriate representation of a geometric object is a fundamental issue for the development of efficient algorithms. Following a recent survey by Kobbelt [15], one may classify the basic types of 3D geometry representations according to the following table Empty Cell Unstructured Structured Hierarchical Explicit Point clouds Binary voxel grid Octree Parametric Triangle mesh NURBS Subdivision surface Implicit Moving least squares surface 3D grid Octree, binary space partitions
Explicit representations
Distance functions
The distance function of a curve or surface M assigns to each point x of the embedding space the shortest distance d(x) of x to M. Since d is not differentiable at M one often uses the signed distance function, which agrees with d up to the sign. It is well defined for a closed object and takes on different signs inside and outside the object, respectively. In the following, we will just speak of the distance function for both the signed and the unsigned version.
The SDM method with the squared distance field attached to the model shape
As input we consider a model shape M. This can be a curve or surface in any analytical or discrete representation (smoothed mesh or a sufficiently dense point cloud with low noise level). The model shape M shall be approximated by a B-spline curve or surface. We will compute a geometric least squares approximant, where distances are measured orthogonal to the model shape M.
For the sake of simplicity in our explanation, we confine ourselves to planar curves, but the concept works for surfaces of
Registration based on squared distance minimization
For the goal of shape inspection, it is of interest to find the optimal Euclidean motion (translation and rotation) that aligns a cloud of measurement points of a workpiece to the CAD model from which it has been manufactured. This makes it possible to check the given workpiece for manufacturing errors and to visualize and classify the deviations. This is one instance of a registration problem. Another registration problem concerns the merging of partially overlapping scans of the same object
Image manifolds for geometry processing
Active curves and surfaces as well as registration problems have their origin in Computer Vision and Image Processing. We would like to point here to another concept which comes from this area and is expected to be of great value for Geometric Modelling and CAD. This is the concept of image manifolds, which has been introduced by Kimmel et al. [14]. Given a 2D image, one associates with each point x=(x,y) in the image plane, an auxiliary point X=(x,y,f1,…,fn) in a higher dimensional space,
Conclusion and future research
Exploiting the huge body of knowledge available in various fields that deal with geometric computing, we can search for unifying methods and in this way simultaneously achieve progress for a number of applications. Even just the adaptation of a method known in one field to an application in another field may lead to remarkable progress. This is a basic philosophy behind Industrial Geometry and has been illustrated at hand of optimization problems involving distance functions and concepts taken
Acknowledgements
Part of this research has been carried out within the Competence Center Advanced Computer Vision and has been funded by the Kplus program. This work was also supported by the Austrian Science Fund under grant P16002-N05 and by the innovative project ‘3D Technology’ of Vienna University of Technology.
Helmut Pottmann is professor of mathematics at the University of Technology in Vienna, Austria. His research interests include classical geometry and its applications, especially to Computer Aided Geometric Design, Computer Vision, industrial geometry, kinematics, and the relations between geometry, numerical analysis and approximation theory. At list of recent publications can be found http://www.geometrie.tuwien.ac.at/pottmann
References (37)
- et al.
Object modeling by registration of multiple range images
Image Vis Comput
(1992) - et al.
Adaptive ruled layers approximation of STL models and multi-axis machining applications for rapid prototyping
J Manufact Syst
(2002) - et al.
Ellipse-offset approach and inclined zig-zag method for multi-axis roughing of ruled surface pockets
Comput-Aided Des
(1998) - et al.
A concept for parametric surface fitting which avoids the parametrization problem
Comput Aided Geom Des
(2003) - et al.
Registration without ICP
Comput Vis Image Understand
(2004) - et al.
Global reparametrization for curve approximation
Comput Aided Geom Des
(1998) Rapid and accurate computation of the distance function using grids
J Comput Phys
(2002)- et al.
Newton methods for parametric surface registration. Part I. Theory
Comput-Aided Des
(2003) - et al.
Reverse engineering
- et al.
Control point adjustment for B-spline curve approximation
Comput-Aided Des
(2004)
An optimal algorithm for approximate nearest neighbor searching
J ACM
A method for registration of 3D shapes
IEEE Trans Pattern Anal Mach Intell
Parametrization-free active contour models with topology control
Vis Comput
Active contours
Fitting subdivision surfaces to unorganized point data using SDM, preprint
Pattern classification
Practical methods of optimization
Adaptively sampled distance fields: a general representation of shape for computer graphics
Comput Graph (SIGGRAPH 00 Proc)
Cited by (0)
Helmut Pottmann is professor of mathematics at the University of Technology in Vienna, Austria. His research interests include classical geometry and its applications, especially to Computer Aided Geometric Design, Computer Vision, industrial geometry, kinematics, and the relations between geometry, numerical analysis and approximation theory. At list of recent publications can be found http://www.geometrie.tuwien.ac.at/pottmann
Stefan Leopoldseder is a member of the mathematics department at the University of Technology in Vienna, Austria. He received his PhD in 1998. His research interests include geometry and its applications to Computer Aided Geometric Design and Computer Vision. Copies of recent publications can be downloaded from http://www.geometrie.tuwien.ac.at/leopoldseder
Michael Hofer is a PhD student at the mathematics department at the University of Technology in Vienna, Austria. He received his master degree in 2000. His research interests include geometric modeling and processing, computer aided geometric design, computer graphics and computer vision. Copies of recent publications can be downloaded from http://www.geometrie.tuwien.ac.at/hofer
Tibor Steiner is researcher at the mathematics department at the University of Technology in Vienna, Austria. He received his PhD in 1993. He has a lot of experience in industrial projects. His research interests include geometry, numerical mathematics and its applications. Copies of recent publications can be downloaded from http://www.geometrie.tuwien.ac.at/tibor
Wenping Wang is Associate Professor of Computer Science at the University of Hong Kong, China. He recieved his BSc and MEng degrees in Computer Science from Shandong University, China, in 1983 and 1986, respectively, and his PhD in Computer Science from University of Alberta, Canada, in 1992. Dr. Wang's research interests include computer graphics, geometric computing, and computational geometry http://www.csis.hku.hk/~wenping/