Paul J. Besl and Neil D.McKay. A method for registration of 3--d shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239 -- 256, February 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....two axes: the number of input views and whether initial pose estimates are known. When a good initial estimate of the relative pose is known, aligning two views is called pair wise registration. The most popular pair wise registration algorithm is the Iterative Closest Point (ICP) algorithm [5]. When more than two views are involved and initial pose estimates are given, the process is called multi view registration. Multi view registration is often used in modeling from reality systems to distribute pair wise registration errors evenly over an entire model. With unknown pose estimates, ....

....minimizing an objective measure of registration error. The dominant algorithm in this category is the iterative closest point (ICP) algorithm, which repeatedly updates the relative pose by minimizing the sum of squared distances between closest points on the two surfaces (point to point matching) [5]. Chen and Medioni proposed a similar method in which the distance between points and tangent planes is minimized instead (point to plane matching) 11] Rusinkiewicz survey of ICP algorithms provides an elegant taxonomy and unifying framework for comparing the numerous extensions to the basic ....

Paul Besl and Neil McKay. A method of registration of 3D shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239--256, February 1992.

....a pre surgical model; discovering this relationship enables, for instance, the realtime overlay of pre surgical planning images on top of images of the actual surgery. The approach taken in [34] is to build an octtree [13] of a model surface, and to then use an iterative closest points algorithm [5] refine an initial alignment estimate to within a very small error. A kd tree is a close relative of an octtree, but is more amenable to spatiallyaware, recursive, parallel, non uniform partitioning; thus, the remainder of this discussion will assume the use of a kd tree in place of an octtree. ....

Paul J. Besl and Neil D. McKay. A method for registration of 3-D shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14:239--256, February 1992.

....a pre surgical model; discovering this relationship enables, for instance, the real time overlay of presurgical planning images on top of images of the actual surgery. The approach taken in [34] is to build an octtree [13] of a model surface, and to then use an iterative closest points algorithm [5] reilne an initial alignment estimate to within a very small error. A kd tree is a close relative of an octtree, but is more amenable to spatially aware, recursive, parallel, nonuniform partitioning; thus, the remainder of this discussion will assume the use of a kd tree in place of an octtree. ....

Paul J. Besl and Neil D. McKay. A method for registration of 3-D shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14:239-256, February 1992.

....on 3D data while the first order and second order features are computed based on 2D data. When the first order and second order features are distorted, the training data samples can become noisy. Therefore, additional information need to be employed. Possible considerations include the blurriness [107] and optic flow [37] or the redundancies 119 in a kinematic limb model [110, 56, 114, 49] However, none of these approaches is without problems, and their computational complexity may prevent from a real time implementation for the time being. Determining Effort qualities from a single camera ....

Alex Pentland. A new sense for depth of field. IEEE Transactions on Pattern Analysis and Machine Intelligence, 9(4):523 531, 1987.

....and M, C, K are respectively the mass, damping, and stiffness matrix of the system. F is the external force that has S attracted by S . On each node of S, the external force F points to the clos est point on S (figure 2) The approach is similar in spirit to Iterative Closest Point methods [2, 22]. Figure 3 illustrates four warping experiments. The first example is the warping of two different faces. The three other outline facial expression, changes in lighting, and out of image plane rotation. c d e f g h Figure 3: XYI warps (original image and warped image) Appearance variations ....

Paul J. Besl and Nell D. McKay. A method for registration of 3D shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239-256, February 1992.

....ffl Given n 4 point correspondances, it is possible to calculate the best rigid displacement (in the least square sense) between the two sets. Two main methods exist: quaternion based [7] and SVD based [1] The availability of such tecniques is at the base of a very popular algorithm: ICP [2]. The best displacement is found assuming correspondance between each point in the rst set and its closest point in the second set. The algorithm consists in applying this transformations and looking for the new closest points in a iterative way. These algorithm works also with other primitives ....

Paul J. Besl and Neil D. McKay. A method for registration of 3-d shapes. IEEE Transactions on pattern Analysis and Machine Inteligence, 14(2), 1992.

....Recently, Greenspan et al. 7] reported experimental evidence that triangle inequality based methods can be superior to the k d tree method, especially for moderate sized point sets. This method was recently applied [6] to improve the performance of the ICP algorithm for registering 3D images [2]. Most existing algorithms for the nearest neighbor problem use spatial partitioning data structures. They are suitable for problems in which the data sets are static and the number of queries is far greater than the size of the data set. However, many applications, such as imaging and graphics, ....

Paul J. Besl and Neil D. McKay. A method for registration of 3D shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239{ 256, 1992.

....nachsten Schritt nun auf einen aktuellen Datensatz ubertragen. Dazu wird zunachst eine starre affin lineare Transformation ausgefuhrt, in der Rotations und Translationsparameter des Modells gegenuber dem aktuellen Datensatz berechnet werden. Hierzu wird das ICP point set matching nach Besl McKay [4] eingesetzt. Dieses Verfahren benotigt als Eingabe zwei Punktmengen und liefert als Ausgabe die Rotationsmatrix und den Translationsvektor, wodurch die Lagebeziehung der beiden Punktmengen beschrieben wird. Glucklicherweise mussen die beiden Punktmengen nicht unbedingt dieselbe Anzahl von ....

Paul J. Besl, Neil. D. McKay. A method for registration of 3-D shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239--256, Feb. 1992.

....a classical problem of computational geometry, and is encountered within computer vision and general pattern recognition. As an example, nearest neighbour determination is the rate determining step in Iterative Closest Point methods, which are ubiquitous within the field of range image processing [1]. We will assume that p c is unique. We also assume that the distance metric is the Euclidean distance and denote it as D ij = jj p i , p j jj. Note that the acceleration methods we propose apply to any metric that satisfies the triangle inequality. For convenience, we denote the distance of any ....

Paul J. Besl and Neil D. McKay. A method for registration of 3d shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239--256, February 1992.

....based on singular value decomposition, and Horn [7, 8] gave closed form solutions using quaternions and orthonormal matrices respectively. If the correspondence between points is not known, or if one of the sets is not a point set, the iterative closest point algorithm (ICP) of Besl and McKay [2] can be used. ICP is a general algorithm for registering a point set to a geometric model, which may be a set of points, lines, or surfaces. Typically the model is constructed preoperatively and the data are gathered intraoperatively. ICP can also be accelerated [2] to speed the iteration in ....

....(ICP) of Besl and McKay [2] can be used. ICP is a general algorithm for registering a point set to a geometric model, which may be a set of points, lines, or surfaces. Typically the model is constructed preoperatively and the data are gathered intraoperatively. ICP can also be accelerated [2] to speed the iteration in regions where the error gradient is smooth. 504 z x y X ray source Image plane Fiducial markers Fig. 1. Imaging ....

[Article contains additional citation context not shown here]

Paul J. Besl and Neil D. McKay. A method for registration of 3-D shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239--256, February 1992.

....vision system as it allows us to focus on the point of fixation. With no focus control, the features that we are fixating on may be out of focus. The ability to control lens focus also allows us to obtain depth information monocularly through focusing [25] or through defocus measurements [22, 29]. Our system also allows control over the lens aperture, which affects the amount of light received by the image sensor, and the depth of focus (not to be confused with the depth of the surface of exact focus) It is important to be able to adjust the aperture to maintain sufficient light levels ....

Alex Pentland. A new sense for depth of field. IEEE Transactions on Pattern Analysis and Machine Intelligence, 9(4):523--531, 1987.

....using bitangent lines or bitangent planes. These are first order semi differential invariants [GMPO92] ffl Then, introducing the normal or tangent, we define a distance between the 3D object and the 2D image, and we minimize it using extensions of the Iterative Closest Point algorithm ( BM92, Zha94] ffl We deal with the critical problem of outliers by computing Mahalanobis distances and performing generalized 2 tests. Results are presented on a variety of real medical data to demonstrate the validity of our approach. Key words: Medical Computer Vision, Registration, Enhanced ....

....bitangents. Ce sont des invariants semi diff erentiels du premier ordre [GMPO92] ffl Puis, en utilisant les tangentes ou les normales, nous d efinissons une distance entre l objet 3D et l image 2D que nous minimisons en utilisant une extension de l algorithme de point le plus proche it eratif ( BM92, Zha94] ffl Nous traitons le probl eme crucial de l occlusion en calculant des distances de Mahalanobis et en r ealisant des tests de 2 g en eralis es. Des r esultats sont pr esent es sur diverses donn ees m edicales pour prouver la validit e de notre approche. Mots cl e : Imagerie ....

[Article contains additional citation context not shown here]

Paul Besl and Neil McKay. A method for registration of 3\GammaD shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239--256, February 1992.

..... 43 8 Conclusion 45 INRIA Randomness and Geometric Features in Computer Vision 5 1 Introduction Many algorithms in computer vision and object recognition deal with simple geome tric features like points, for example the Iterative Closest Point ([BM92, Zha94]) the geometric hashing ( LW88, RH93] and the alignment ( AF86, HU87] algorithms. On the other hand, models of the real world often lead to the consideration of more complex features: lines ( Gri92] planes, oriented points, frames ( PA94, PT95b] etc. The handling of these features ....

....the left and right invariant distances are obviously dioeerent. This reAEects the fact that N (f 1 ffi f 2 ) 6= N (f 2 ffi f 1 ) 5. 6 Utility of invariant distances The distance between two points is often used in geometric algorithms: the Iterative Closest Point algorithm, developed in [BM92, Zha94] is the best example. Another example is the classic method of computing the transformation that maps a set of points x i in one image, to a set of points y i in another image: it is the transformation which the minimizes the sum of squared distances C(f) P i dist 2 (f x i ; y i ) All ....

Paul Besl and Neil McKay. A method for registration of 3\GammaD shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239256, February 1992.

....M, C, K are respectively the mass, damping, and stiffness matrix of the system. F is the external force that has S attracted by S 0 . On each node of S, the external force F points to the closest point on S 0 (figure 2) The approach is similar in spirit to Iterative Closest Point methods [2, 22]. Figure 3 illustrates four warping experiments. The first example is the warping of two different faces. The three other outline facial expression, changes in lighting, and out of image plane rotation. x I(x) S S Figure 2: 1D representation of S being pulled towards S 0 a b c d e f g ....

Paul J. Besl and Neil D. McKay. A method for registration of 3D shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239--256, February 1992.

.... retrieved from database, from [VV99] 2 Approaches Matching has been approached in a number of ways, including tree pruning [Ume93] the generalized Hough transform or pose clustering [Bal81] Sto87] geometric hashing [WR97] the alignment method [HU87] statistics [Sma96] deformable templates [SP95] relaxation labeling [RR80] Fourier descriptors [Lon98] wavelet transform [JFS95] curvature scale space [MAK96] and neural networks [Gol95] Without being complete, in the following subsections we will describe and group a number of these methods together. 2.1 Global image transforms There ....

Stan Sclaro and Alex P. Pentland. Modal matching for correspondence and recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 17(6), June 1995.

....fea tures, and (3) what is the imean featurej of a num ber of feature measurements, and we propose generic methods to solve them. 1 Introduction Many algorithms in computer vision and object recog nition deal with simple geometric features like points, for example the Iterative Closest Point ([4, 21]) the ge ometric hashing ( 13, 19] and the alignment ( 3, 11] algorithms. On the other hand, models of the real world often lead to the consideration of more complex features: lines ( 8] planes, oriented points, frames ( 16, 18] etc. The handling of these features raises some problems ....

....(and thus left and right invariant measures are identical) This reAEects the fact that N (f 1 ffi f 2 ) 6= N (f 2 ffi f 1 ) 4. 6 Utility of invariant distances The distance between two points is often used in ge ometric algorithms: the Iterative Closest Point algo rithm, developed in [4, 21] is the best example. An other example is the classic method of computing the transformation that maps a set of points x i in one image, to a set of points y i in another image: it is the transformation which the minimizes the sum of squared distances C(f) P i dist 2 (f x i ; y i ) All ....

Paul Besl and Neil McKay. A method for registration of 3\GammaD shapes. IEEE Transactions on Pattern Analy sis and Machine Intelligence, 14(2):239256, February 1992.

....to be polyhedral, piecewisequadric or superquadric; the shape of the object can be arbitrary. Recent approaches using algebraic polynomials [10, 16] splash and super (polygonal) segments [17] simplex angle image [4] 2D silhouettes with internal edges [3] and point sets based registration [2] have specifically sought to address the issue of representing complex curved free form surfaces. They suffer from one or more limitations relating to segmentation issues, surface fitting convergence, bounding constraints, restricting objects to be topologically equivalent to a sphere, local ....

Paul J. Besl and Neil D. Mckay. A method for registration of 3-D shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239--256, 1992.

.... Robotique, image et vision Projet Epidaure Rapport de recherche n2220 March 1994 36 pages Abstract: In this paper, we propose a new framework to perform nonrigid surface registration. It is based on various extensions of an iterative algorithm recently presented by several researchers ([BM92], Zha93] CM92] ML92] CLSB92] to rigidly register surfaces represented by a set of 3D points, when a prior estimate of the displacement is available. Our framework consists of three stages: ffl First, we search for the best rigid displacement to superpose the two surfaces. We show how to ....

....et localement affine de surfaces gauches R esum e : Dans cet article, nous proposons une nouvelle approche pour r ealiser le recalage non rigide de surfaces discr etes. Cette approche est fond ee sur diverses extensions d un algorithme it eratif r ecemment pr esent e par plusieurs chercheurs ([BM92], Zha93] CM92] ML92] CLSB92] pour recaler rigidement des surfaces repr esent ees par des ensembles de points, lorsque l on dispose d une estim ee du d eplacement cherch e. Elle consiste en trois etapes : ffl Premi erement, nous cherchons la meilleure transformation rigide pour superposer ....

[Article contains additional citation context not shown here]

Paul Besl and Neil McKay. A method for registration of 3\GammaD shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239--256, February 1992.

....be very simple if to each point of S 1 , we knew in advance the point in S 2 which corresponds, but this is one of the unknowns of the problem. Therefore, an iterative algorithm composed of two steps is often used to find the minimizing transform (see ICP, Iterative Closest Point algorithm in [31, 32, 33]) Step 1 : Matching. For a previous estimate of the transform (R n ; t n ) find the matching between points of the transformed shape (R n S 1 t n ) and S 2 . This is done finding for each X 1 i in S 1 the closest point X 2 jn (i) in S 2 to (R n X 1 i t n ) The output is a ....

....model. 7.2 Auxiliary variable formulation In the case the distance d in Eqn. 83) is the same as the one induced by the norm in Step 2, this algorithm ensures convergence since at each step it is the same term that is minimized with respect to each of the variables (R; t) or j n as shown in [31]. However, in the case d is a distance in a higher dimension space, including other features like in [33] or in the case the potential is a function of the distance taking into account only the best matches like in robust estimation, there is no reason for convergence and our formulation gives a ....

Paul Besl and Neil McKay. A method for registration of 3\GammaD shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239--256, February 1992.

....processes, which involve the evaluation of a geometric transform. The aim of matching methods is generally to reduce the complexity of the feature association. See, for example, AF86] or [HU87] for Alignment, Gri92] for Interpretation Trees, LW88, Wol90] or [RH93] for Geometric Hashing, [BM92] or [Zha94] for Iterative Closest Point (ICP) methods. In the following, we do not discuss matching methods per se, rather the estimation of the transform. The traditional approach is to apply a least squares method, using for example the singular value decomposition (see [AHB87] Ume91] or the ....

Paul Besl and Neil McKay. A method for registration of 3\GammaD shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239--256, February 1992.

....of objects known to be in the scene is accurately recovered, i.e. the parameters of the rigid displacement between each object model and its partial reconstruction obtained through image analysis are computed. Currently the algorithms used are adapted from the Iterative Closest Point approach [BM92, V ez95] see Figures 8c and 4) 3D reconstruction Object registred Figure 4: The 3D steps of analysis process. Simplifying assumptions are made during the analysis process which constrain the possible choices in the observed scene as well as in the image sequences: ffl Camera: camera moves ....

Paul J. Besl and Neil D. McKay. A method for registration of 3D shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239--256, February 1992.

.... (e.g. stereo vision) Two basic passive approaches can be distinguished according to the way, how the input data is interpreted: i) geometric stereo [8] using multiple views projections 1 Introduction of the same 3D structure, and (ii) shape from X (shading [20, 22, 27] texture [26] defocus [29], motion[25] Moir e maps [37] etc. using typically image from a single vantage point . The both approaches combination is possible [9, 11] There are two main kinds of output model representations: volume based and surface based . Volume based models (e.g. generalized cylinders, quadrics or ....

Alex Paul Pentland. A new sense for depth of field. IEEE Transactions on Pattern Analysis and Machine Intelligence, 9(4):523--531, July 1987.

....different lens parameters are playing the active role. For in focus methods it is (almost) only the focus setting that is controlled, while defocus algorithms typically rely on changing the aperture. One of the first to use image sharpness as a depth cue in computer vision systems was Pentland [Pentland, 1987]. His approach belongs to the category of depth from defocus, where measuring the image blur is essential. High accuracy is in general difficult to obtain with this technique since it is highly dependent on the sensing apparatus as well as the underlying structure of the scenario. Pentland ....

Alex Paul Pentland. A new sense for depth of field. IEEE Transactions on Pattern Analysis and Machine Intelligence, 9(4):523--531, July 1987.

No context found.

Paul J. Besl and Neil D.McKay. A method for registration of 3--d shapes. IEEE Transactions on Pattern Analysis and Machine Intelligence, 14(2):239 -- 256, February 1992.

No context found.

Alex P. Pentland and Bradley Horowitz. Recovery of nonridid motion and structure. IEEE Transaction on Pattern Analysis and Machine Intelligence, 13(7):730--742, 1991.

No context found.

Paul Besl and Neil McKay. A method for registration of 3\GammaD shapes. IEEE Transactions on Pattern Analy sis and Machine Intelligence, 14(2):239256, February 1992.

No context found.

Alex P. Pentland and Bradley Horowitz. Recovery of nonridid motion and structure. IEEE Transaction on Pattern Analysis and Machine Intelligence, 13(7):730--742, 1991.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC