## Recovery of Parametric Models from Range Images: The Case for Superquadrics with Global Deformations (1990)

Venue: | IEEE Transactions on Pattern Analysis and Machine Intelligence |

Citations: | 189 - 5 self |

### BibTeX

@ARTICLE{Solina90recoveryof,

author = {Franc Solina and Ruzena Bajcsy and Arqd Ruzena Bajcsy and Senior Member},

title = {Recovery of Parametric Models from Range Images: The Case for Superquadrics with Global Deformations},

journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},

year = {1990},

volume = {12},

pages = {131--147}

}

### Years of Citing Articles

### OpenURL

### Abstract

this paper, we introduce a method for recovery of compact volumetric models for single part objects. To solve the shape recovery problem in isolation from segmentation, we assume that only a single object is present in the scene at a time. Although we made this simplification to break up the problem, this assumption is still valid for some restricted environments [30]. We show that the shape of those objects can be recovered subject to the model's internal constraints. In this work we use a partic- ular example of compact volumetric models--superquad- ric primitives with parametric deformations. We introduce a least-squares minimization method to recover model and deformation parameters using range data as the input. Range data enables us to study shape recovery independent of different passive techniques of obtaining depth data, such as depth from stereo, depth from focus, or depth from motion. The fitting function which we min- imize is a cost or energy function whose value depends on the distance of points from the model's surface and on the overall size of the model. We show that the solution space, which can have more than one "deep" minimum or acceptable solution and many shallow local minima, can be searched efficiently with a gradient descent method. By using a stochastic technique, the procedure can escape from shallow local minima, and a particular solution among several acceptable solutions can be reached by searching in a constrained parameter subspace. The paper is organized as follows. Section II is on parametric models in computer vision, focusing on comparison of generalized cylinders and superquadrics. Section III explains superquadric models in detail. Section IV is about recovery of nondeformed superquadric models, and Section V is on recovery of defo...

### Citations

1406 |
Robot vision
- Horn
- 1986
(Show Context)
Citation Context ...+ (zi - -(zi - - y) . (xi - + (yi (9) Central moments are moments with respect to the center of gravity (, y,s). We want to find a rotation matrix R which makes the matrix of moments M diagonal [28], =-=[16]-=-. The new diagonal matrix of moments D is then O -- -1i, (10) where R is the rotational part of transform T in Section HI-B. On the other hand, matrix M can be diagonalized with a diagonalization matr... |

834 |
Modularity of mind
- Fodor
- 1983
(Show Context)
Citation Context ...us with information that makes it possible to locate and recognize objects and their interrelationships without direct physical contact. In short, vision makes the outside world accessible to thought =-=[11]-=-. Although we are far from understanding the complexities of human visual perception, we are trying to give machines the sense of vision for the very same reason--to enable them to interact with a cha... |

829 | Computer Vision - Ballard, Brown - 1982 |

720 | Elastically Deformable Models
- TERZOPOULOS, PLATT, et al.
- 1987
(Show Context)
Citation Context ... Fig. 19. Model, recovered for an irregular object--a tennis shoe. could be enhanced also with parameters representing some physical properties related to the real world, like elasticity or stiffness =-=[32]-=-. What seems to be important is the chosen granularity of representation. There is psychological evidence showing the particular salience of parts--especially on the basic category level [35J. For obj... |

705 |
Numerical Recipes
- Press, Teukolsky, et al.
- 1993
(Show Context)
Citation Context ...l stability. For example, when any of parameters a, a2, a3, q, e2 = 0, the inside-outside function (3) becomes singular. We use the Levenberg-Marquardt method for nonlinear least squares minimization =-=[25]-=-, [29] since first derivatives bR/bai for i = 1,s, 11 can be computed analytically. Parameter constraints are implemented by a simple projection method [29]. We take the search vector or the trial set... |

704 |
Linear Algebra and its Applications
- Strang
- 1988
(Show Context)
Citation Context ...-- -1i, (10) where R is the rotational part of transform T in Section HI-B. On the other hand, matrix M can be diagonalized with a diagonalization matrix , whose columns are eigertvectors of matrix M =-=[31]-=- V = Q 'MQ. (11) makes up the rotational part of transform T--(4)] we compute the equivalent Euler angles qt, 0t, ;bE [22]. For evaluating the inside-outside function (5) we could use the elements of ... |

312 |
Global and local deformations of solid primitives
- Barr
- 1984
(Show Context)
Citation Context ...Of interest are deformations that occur often in nature or are used for manufacturing of man-made objects. Deformations such as simplified tapering, bending and twisting require just a few parameters =-=[5]-=-. A shape deformation is a function D which explicitly modifies the global coordinates of points in space X = D(x) (13) where x are the points of the undeformed solid and X are the corresponding point... |

290 |
Multiresolution elastic matching
- Bajcsy, KovaEiE
(Show Context)
Citation Context ...del gives a good starting point for recovery of local surface patch models. For refinement of shape representation, local deformations could be used by applying local push and pull to the part models =-=[3]-=-. Models SOLiNA AND BAJCSY: RECOVERY OF PARAMETRIC MODELS FROM RANGE IMAGES 145 Fig. 19. Model, recovered for an irregular object--a tennis shoe. could be enhanced also with parameters representing so... |

218 |
Internal representation of Solid Shape with respect to Vision
- Koenderink, Doorn
- 1979
(Show Context)
Citation Context ...re visible at the same time. We assume a general view for recovery of object shapes. Seeing, for example, just one side of a cube does not provide enough information on the extent of the whole object =-=[18]-=-, [17]. But even assuming a general viewpoint, objects such as parallelepipeds or cylinders (objects with surfaces where at least one principal curvature = 0) do not provide enough constraints lbr sha... |

158 |
Perceptual organization and the representation of natural form
- Pentland
- 1986
(Show Context)
Citation Context ...lity in shape description for a simpler and more efficient recovery of models by adopting a shape vocabulary for modeling on the level of parts which consists of superquadrics and global deformations =-=[23]-=-. Description of shapes on the pan level does not require complicated models. While a cylinder with a bump on it could be modeled with a single primitive in the modeling system proposed by Terzopoulos... |

145 | Model-based recognition and localization from sparse range or tactile data
- Grimson, Lozano-Perez
- 1984
(Show Context)
Citation Context ...Most so called "model driven" object recognition vision systems use context dependent information by relying on a set of rigid or precise models for all objects expected to be found in the s=-=cene [7], [14]-=-. This approach is possible in tightly controlled environments, such as in industrial applications. It clearly cannot work in unconstrained environments due to the sheer number of necessary models, no... |

114 | Symmetry-Seeking Models and 3D Object Reconstruction
- Terzopoulos, Witkin, et al.
- 1987
(Show Context)
Citation Context ...sic forces are the influences which direct the shape options allowed by the internal constraints. Shape is a result of the interaction of intrinsic and extrinsic forces. Terzopoulos, Witkin, and Kass =-=[33]-=- have proposed a 3-D shape model similar to generalized cylinders, but enhanced with deformation parameters to control the elasticity of the main axis and the walls of the cylinder. Using this powerfu... |

102 |
On the role of structure in vision
- Witkin, Tenenbaum
- 1983
(Show Context)
Citation Context ...on compression by infusion of outside knowlege is required. This can be done either by retaining the low level models and selecting only the relevant ones by using the laws of perceptual organization =-=[37]-=-, or by building compact volumetric models of larger granularity. In either case, if the additional information is not explicit, this is a difficult step to make. The additional knowledge should enabl... |

73 |
Introduction to nonlinear optimization
- Scales
- 1985
(Show Context)
Citation Context ...ility. For example, when any of parameters a, a2, a3, q, e2 = 0, the inside-outside function (3) becomes singular. We use the Levenberg-Marquardt method for nonlinear least squares minimization [25], =-=[29]-=- since first derivatives bR/bai for i = 1,s, 11 can be computed analytically. Parameter constraints are implemented by a simple projection method [29]. We take the search vector or the trial set of pa... |

70 | Energy constraints on parameterized models
- Witkin, Fleischer, et al.
- 1987
(Show Context)
Citation Context ...ding points after deformation. Both x and X are expressed in the object centered coordinate system. Any translation or rotation is performed after the deformation. This can be described schematically =-=[38]-=- by Trans (Rot (Deform (x))). (14) Tangent and normal vectors at every point on the deformed surface, which are important for rendering or for checking the consistency of the model with the input rang... |

65 | Learning physical descriptions from functional definitions, examples, and precedents
- Winston, Binford, et al.
- 1983
(Show Context)
Citation Context ...rate parts have separate functions, similar parts have similar functions, and more salient parts have more important functions. This close relation opens up possibilities to infer function from shape =-=[36]-=-. There is also ample evidence that human perception makes use of the prototype and deformation paradigm. Using deformations is an efficient way of representing a large variety of shapes and has a cau... |

19 |
The superellipse: a curve that lies between the ellipse and the rectangle. Scientific American
- Gardiner
- 1965
(Show Context)
Citation Context ...ng several hypothesis select the simplest--which guides human perception [13]. III. SUPERQUADRICS Superquadrics are a family of parametric shapes that were discovered by the Danish designer Peit Hein =-=[12]-=- as an extension of basic quadric surfaces and solids. Superquadrics have been used or proposed for use as primitives for shape representation in computer graphics [4] and computer vision [23]. Superq... |

16 |
Recovery of superquadrics from depth information
- E, Gross
- 1987
(Show Context)
Citation Context ...el = 2, e: = 1, a3 >> al, a2), even when the initial estimates are close to the desired solution. The same problem is described elsewhere although no explanation for the cause of the problem is given =-=[8]-=-. B. Superquadric Inside-Outside Function for General Position and Orientation The inside-outside function (3) defines the superquadric surface in an object centered coordinate system (Xs, Ys, zs). In... |

9 | Computing volume descriptions from sparse 3-D data
- Rao, Nevatia
- 1988
(Show Context)
Citation Context ...resentations, using rule base reasoning, ACRONYM being the most representative of such systems [10]. Generalized cylinders can also be recovered directly from short edge segments obtained from stereo =-=[26]-=-. Common to all these model recovery methods is that they are based on rules, such as the importance of parallel lines, which are derived from perceptual organization ideas. In such rule-based systems... |

8 |
Nested structures of control: An intuitive view
- Leyton
- 1987
(Show Context)
Citation Context ...s. Tapering usually precedes bending, or in other words, tapering is perceived as to affect the model along the longest axis of symmetry for elongated objects, whether the axis is bent or not. Leyton =-=[19]-=- has shown that deformations acting on prototypical shapes have a specific order, such that the transformation is more structure-preserving. The corresponding model has the following structure Trans (... |

8 |
Principles of Categorization" in Cognition and Categorization (Ed.s
- Rosch
- 1978
(Show Context)
Citation Context ...ers are perceptually relevant. The adopted shape vocabulary is intended for rough description of parts, suitable for object avoidance, grasping or shape classification/recognition of basic categories =-=[27]-=-. Inadequate representation can occur if thc occluded side of an object is not symmetrical to the visible side, or because the deformations that we use are defined only along the longest side of the o... |

4 |
Range image interpretation of mail pieces with superquadrics
- Solina, Bajcsy
- 1987
(Show Context)
Citation Context ...n, we assume that only a single object is present in the scene at a time. Although we made this simplification to break up the problem, this assumption is still valid for some restricted environments =-=[30]-=-. We show that the shape of those objects can be recovered subject to the model's internal constraints. In this work we use a particular example of compact volumetric models--superquadric primitives w... |

1 |
Three dimensional shape repreeutarion revisited
- Bajcsy, Solina
- 1987
(Show Context)
Citation Context ...s on the inside of the cavity have the wrong sign. In this way, a cavity can be inferred. To recover cavities, we could use the surface normal information during model recovery as we suggested before =-=[2]-=-. We decided to do it in two steps instead (Fig. 8). First we use the fitting function (7) to recover a convex model enclosing the object. If the surface normals through the input range points are not... |

1 |
Superquadrics and angle-preserviug transformations," IEEEComput
- Ban-
- 1981
(Show Context)
Citation Context ...y the Danish designer Peit Hein [12] as an extension of basic quadric surfaces and solids. Superquadrics have been used or proposed for use as primitives for shape representation in computer graphics =-=[4]-=- and computer vision [23]. Superquadrics play the role of prototypical pans and can be further deformed and glued together into realistic looking models as is nicely demonstrated by Pentland's Supersk... |

1 |
Human image understanding: Receut research aud theory
- Biederman
- 1985
(Show Context)
Citation Context ..., it would require two primitives or pans in the vocabulary that we use--a cylinder and a bump. This level of description is also favored by some psychologists when trying to explain human perception =-=[6]-=-. Superquadric models are implicitly constrained by the class of shapes that they can model and do not require any setting of external constraints from case to case. An important advantage for ease of... |

1 |
3DPO: A three-dimensmnal part orientation system
- Bolles, Horaud
- 1986
(Show Context)
Citation Context ...dge. Most so called "model driven" object recognition vision systems use context dependent information by relying on a set of rigid or precise models for all objects expected to be found in =-=the scene [7]-=-, [14]. This approach is possible in tightly controlled environments, such as in industrial applications. It clearly cannot work in unconstrained environments due to the sheer number of necessary mode... |

1 |
Multi-level adaptive solutions to boundary value problems
- Bran&
- 1977
(Show Context)
Citation Context ...ly vision algorithms [1]. One can view the coarser grids as correction grids, accelerating convergence of the minimization scheme on the finest grid by efficiently liquidating smooth error components =-=[9]-=-. Fast and efficient recovery of superquadric models can be done on a hierarchy of range images with different resolution. We implemented a multiresolution model recovery scheme which starts on a very... |

1 |
Category induction and representation," in Categorical Perception
- Hamad
- 1986
(Show Context)
Citation Context ...thin-category parameter differences look smaller than between-category parameter differences even when they are of the same size. This differential responding is referred to as categorical perception =-=[15]-=-. By mapping symbols on the continuous model parameter space it is possible to define distinctive volumetric primitives such as Geons, a finite set of shape primitives believed to correspond to letter... |

1 |
Kender and D. Freudensteim "What is a 'degenerate view
- R
- 1987
(Show Context)
Citation Context ...ible at the same time. We assume a general view for recovery of object shapes. Seeing, for example, just one side of a cube does not provide enough information on the extent of the whole object [18], =-=[17]-=-. But even assuming a general viewpoint, objects such as parallelepipeds or cylinders (objects with surfaces where at least one principal curvature = 0) do not provide enough constraints lbr shape rec... |

1 |
Cognitive constructs on ordering operations: The case of geometric analogies
- Novick, Tversky
- 1987
(Show Context)
Citation Context ...shape. This order resembles the structure of solving geometry analogy tasks which involve finding first location, orientation, size, general outline, and additional small parts and details at the end =-=[21]-=-. The model parameter space (superquadrics with deformations) is continuous, but some sets of parameters correspond to easily identifiable geometric primitives such as parallelepipeds, ellipsoids and ... |

1 |
Segmentation of 3-D scenes using multi-modal interaction between machine vision and programmable mechanical scene manipulation
- Tsikos
- 1987
(Show Context)
Citation Context ...ntation and size of the model (Section IV-B), which is necessary to initialize minimization. Throughout this work we used range images obtained with a laser imager built at University of Pennsylvania =-=[34]-=-. Before model recovery, the supporting surface was removed by fitting a plane to it and subtracting the points on or close to that plane. A. Least-Squares Minimization of Superquadric Inside Outside ... |