D. W. Thompson and J. L. Mundy. Three-dimensional model matching from an unconstrained viewpoint. In Proceedings of the IEEE Conference on Robotics and Automation, pages 208--220, 1987.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....ORA system of Huttenlocher [7] and the geometric hashing system of Damdan et al. 9] have relied on the computation of the pose of triangles by weak perspective approximations. Other object recognition systems have also used weak perspective approximation on nontriangular combinations of features [14], 15] However, the authors do not discuss the errors introduced by the approximation in comparison with an exact perspective approach. This is one of the topics of this paper. We also show that weak perspective is only one possible approximation for simplifying the problem and is not necessarily ....

D.W. Thompson and J. L. Mundy, "Three-dimensional model matching from an unconstrained viewpoint," in Proc. IEEE Robotics Automat., 1987, pp. 208-220.

.... and image features (e.g. Clark et al. 13] Fischler and Bolles [17] Ayache and Faugeras [5] Horaud [25] Huttenlocher and Ullman [27] Other approaches use indexing to match more than the minimal number before looking for confirming features (e.g. Rothwell et al. 38] Thompson and Mundy [43], Lamdan et al. 32] Jacobs [29] Most recognition systems take an ad hoc approach to the problem of accounting for the effects of sensing error on the projected positions of unmatched model features. Some systems match projected model features to image features if they are separated by a ....

....system might attempt to add matches, and use these additional matches to narrow the area in which it must search for still more consistent matches. Additionally, the algorithm from Section 6 may be useful in methods that match image to model features by indexing, and then verify these matches [32, 14, 29, 43, 38, 45]. In these approaches, some model features are matched to image features to determine a model pose, and then this pose is used to find matches for additional model features. Our results show exactly where to search for these matches when we have matched three image and model points. As mentioned ....

Thompson, D., and J. L. Mundy, "Three-Dimensional Model Matching From an Unconstrained Viewpoint", Proc. IEEE Conf. Rob. Aut., pp. 208-220, 1987.

....finely than one would like. Each dimension of that table was divided into 40 parts, compared to 100 parts in the current system. Also in general, space requirements grow as group size grows. Finally, building the lookup table with sampling undoubtedly resulted in many table entries being missed. [23] had previously used a similarly constructed lookup table to determine the object pose implied by a match between a pair of image vertices and a pair of model vertices. They also built their table by sampling the images a model produces, doing this by sampling the viewing sphere. Their table ....

Thompson, D. and Mundy, J., 1987. "Three-Dimensional Model Matching from an Unconstrained Viewpoint." IEEE Proc. Robotics and Automation:208-220.

....or edges) to represent objects. An object is recognized in this approach if there exists a viewpointfrom which the model features coincide with the corresponding image features, e.g. Roberts, 1965, Fischler and Bolles, 1981,Lowe, 1985, Huttenlocher and Ullman, 1987, Basri and Ullman, 1988, Thompson and Mundy,1987, Ullman and Basri, 1991] Since images often are noisy and models occasionally are imperfect, it is rarely the case that a model aligns perfectly with the image. Systems therefore look for a model that reasonably aligns with the image. Consequently, measures that assess the quality of a match ....

Thompson, D. W. and Mundy, J. L. (1987). Three-dimensional model matching from an unconstrained viewpoint. In Proceedings of IEEE Conferenceon Robotics and Automation, pages 208--220, Raleigh, NC.

....many of the techniques for recognizing objects by part decomposition rely on finding the entire parts fi om the image. To recognize the specific identity of objects, a relatively detailed representation of the object s shape is compared with the image. An example for such meth ods is alignment [3, 9, 12, 13, 18, 25, 40, 41]. Alignment involves recovering the position and orientation (pose) in which the object is observed and comparing the appearance of the object fi om that pose with the image. Only a few attempts have been made in the past to extend the alignment scheme to the problem of object catego rization ....

Thompson, D.W. and Mundy J.L., 1987. Three di- mensional model matching from an unconstrained viewpoint. Proc. of lEEE Int. Conf. on robotics and Automation, 208-220.

.... (e.g. 5] 10] 1] 9] 28] 29] 15] 3] 16] 18] 30] 19] In addition, pose clustering techniques use every correspondence between a minimal set of model and image features to compute a model pose, and then count the number of times each pose is repeated (e.g. 2] 26] [25], 23] 11] 4] For computing poses of 3D objects from 2D images, a model of projection must be selected, and typically either perspectiveor weak perspective projection is chosen. Weak perspective projection is an orthographic projection plus a scaling, whichserves to approximate perspective ....

....model points just computed. A simple method for doing so is given in Appendix A# for a least squares solution, see Horn [14] Although perspective(central) projection is a more accurate model, numerous researchers have used weak perspective projection instead (e.g. 24] 20] 7] 8] [25], 28] 29] 21] 22] 16] 18] 3] 30] 19] 12] The justification for using weak perspectiveisthatinmany cases it approximates perspective closely. In particular, for many imaging situations if the size of the model in depth (distance in z) is small compared to the depth of the ....

[Article contains additional citation context not shown here]

Thompson, D. W., and J. L. Mundy, "Three-Dimensional Model Matching from an Unconstrained Viewpoint," in Proc. IEEE Conf. Rob. Aut., pp. 208-220, 1987.

....[1] Besl and Jain [2] A number of working recognition schemes have been proposed for simple 3D objects. Lowe [3] presents a 3D object recognition system which uses a single image, pereeptual groupings and viewpoint consistency constraints to detect 3D objects from 2D data. Thompson and Mundy [4] use vertex pairs to derive the arline transformation between a 3D polyhedral object model and its projection into the image viewplane. A more common approach is the use of 3D data extracted from range data or multiple views. The approach by Grimson and Perez [5] operates by examining all ....

D.W. Thompson and J.L. Mundy. Three-dimensional model matching from an unconstrained viewpoint. In Proceedings of IEEE Coferece o Robotics ad Automations, pages 208 220, 1987.

....of features in the image with the corresponding features in the 3D model. The aligning transformation is computed separately for each of the models stored in the system. The outcome of the recognition process is the model that fits the input most closely after the two are aligned. Related schemes [14,29] choose the best model using viewpoint consistency constraints, which relate the projected locations of the features of a model to its 3D structure, given a hypothesized viewpoint. Three dimensional models are also postulated by those recognition theories that represent objects by 3D structural ....

D. W. Thompson and J. L. Mundy. Three-dimensional model matching from an un- constrained viewpoint. In P'oceedigs of IEEE Cofe'ece o Robotics ad Automations, pages 208 220, Raleigh, NC, 1987.

....distance of the object s points from the image plane. Image scaling is used to model the effect of object to camera distance on the object s projection; it is a good approximation to perspective projection when the camera s distance to the object is much larger than the size of the object itself [36]. c) The image projection of an affinely represented point p is given by [p T p T ] T , where and are the directions of the rows and columns of the camera, respectively, in the reference frame of the affine basis points. The camera s internal reference frame is defined by the vectors ....

....errors comparable. Results are shown in Figs. 15 and 16. While errors remain within 15 pixels for the range of motions we considered (in a 640 Theta 480 image) the results show that, as expected, the affine approximation to perspective leads to errors as the distance to the object decreases [36, 49]. These effects suggest the utility of projectively invariant representations for representing virtual objects when the object camera distance is small. The accuracy of the real time video overlays generated by our system was measured as follows. A pair of region trackers was used to track the ....

[Article contains additional citation context not shown here]

W. B. Thompson and J. L. Mundy, "Three-dimensional model matching from an unconstrained viewpoint," in Proc. IEEE Robotics Automat. Conf., pp. 208--220, 1987.

.... and Clemens[14] for example, determine pose based on a match between line segments, while Fischler and Bolles[18] Huttenlocher and Ullman[26] Horaud[22] Ullman and Basri[57] Jacobs[27] Rothwell et al. 47] and Alter and Jacobs[1] use point features to determine pose, and Thompson and Mundy[54] make use of vertices. It is fairly well understood how to use local features for pose determination or indexing, but they have significant weaknesses. Local features often do not capture the shape of complex, curved 3 D objects. And it may be quite difficult to locate 2 D image features that ....

Thompson, D. and Mundy, J. 1987. Three-Dimensional Model Matching From an Unconstrained Viewpoint. In Proceedings IEEE Conference Rob. Aut., pp. 208-220.

....a linear programming algorithm to compute this uncertainty when poses are based on any number of initial matches. 1 Introduction Object recognition systems frequently hypothesize a known object s pose based on matching a small number of the object s features to features in the image (e.g. [7, 14, 17, 10]) To confirm the hypothesis, they commonly use the pose to look for additional matches. A fundamental question in building robust recognition systems is how noise in the matched image features can propagate to uncertainty in the locations of still unmatched object features. Here we consider ....

.... In general, we can place a convex polygon of any shape about the possible locations of the n 1 st point [12] This algorithm could be quite useful to alignmenttype approaches to recognition [7, 14, 10] or in methods that match image to model features by indexing and then verify these matches [17]. A complete recognition system using this result might work as follows: 1. Match k image and model points, using a search or indexing method. Assume that the error in each image point is bounded by some m sided polygon. 2. Use the matches to generate km linear constraints on the possible errors ....

Thompson, D. W., & J. L. Mundy, "Three-Dimensional Model Matching from an Unconstrained Viewpoint," Proc. IEEE Conf. Rob. Aut., 1987.

....the solutions using the affine projection model and the perspective model are negligible. As a rule of the thumb the ratio between the distance of the object from the camera and the depth differences of the individual features should be at least 10:1 [ Costall, 1993 ] Gee and Cipolla, 1994 ] Thompson and Mundy, 1987 ] Algorithms proposed to solve the three point to 3Dpose problem include Ullman s [ Ullman, 1986 ] Huttenlocher and Ullman s [ Huttenlocher and Ullman, 1990 ] Grimson, Huttenlocher and Alter s [ Grimson et al. 1992 ] Alter s [ Alter, 92 ] and Cygnaski and Orr s [ Cyganski and Orr, 1988 ] ....

D. W. Thompson and J. L. Mundy. Three-dimensional model matching from an unconstrained viewpoint. In Proc. IEEE Robotics and Automation, pages 208--220, April 1987.

.... [14] for example, determine pose based on a match between line segments, while Fischler and Bolles [18] Huttenlocher and Ullman [26] Horaud [22] Ullman and Basri [58] Jacobs [27] Rothwell et al. 47] and Alter and Jacobs [1] use point features to determine pose, and Thompson and Mundy [55] make use of vertices. It is fairly well understood how to use local features for pose determination or indexing, but they have significant weaknesses. Local features often do not capture the shape of complex, curved 3 D objects. And it may be quite difficult to locate 2 D image features that ....

D. Thompson and J. Mundy, 1987. "Three-Dimensional Model Matching From an Unconstrained Viewpoint," In Proc. IEEE Conference on Robotics and Automation: 208--220.

....in the scene. A common approach to handling the problem of recognition from different viewpoints is by comparing the stored models to the observed environment after the viewpoint is recovered and compensated for. This approach, called alignment, is used in a number of studies of object recognition [3, 8, 10, 13, 18, 19]. We apply the alignment approach to the problem of localization. Below we describe a localization system based on the Linear Combinations scheme [20] The presentation is divided into two parts. In the first part (Section 2.1) we describe the basic system that works under weak perspective ....

D. W. Thompson and J. L. Mundy. Three dimensional model matching from an unconstrained viewpoint. Proc. Int. Conf. on Robotics and Automation, Raleigh, NC, pp. 208-- 220, 1987.

....process. First, the position and orientation (pose) of the observed object is recovered, and then the model is transformed to this pose, projected to the image plane, and compared with the actual image. A large number of studies use alignment like algorithms to recognize 3D object from 2D images [5, 6, 7, 8, 35, 36, 37]. These studies vary in the representations used and the method employed to recover the alignment transformation. Most of these Viewer Centered Representations: a Computational approach 5 studies use object centered representions. When viewer centered representations are used, the naive approach ....

....transformation. Most of these Viewer Centered Representations: a Computational approach 5 studies use object centered representions. When viewer centered representations are used, the naive approach usually is taken; namely, the system can recognize only the store views of an object (e.g. [37, 38, 39]) For example, in [37] an object is modeled by a large number of views (the representation includes a table of 72 Theta 72 = 5184 views) A view is recognized only if the image is related to one of these views by a rotation in the image plane, in which case this view and the image share the same ....

[Article contains additional citation context not shown here]

D. W. Thompson and J. L. Mundy. Three dimensional model matching from an unconstrained viewpoint. In Proceedings of IEEE International Conference on robotics and Automation, pages 208--220, Raleigh, NC, 1987.

....the 3 x 4 homogeneous projection matrix, T. We will refer to projections characterized by a 3 x 4 homogeneous matrix as a camera. Perhaps the most widely used form in vision literature is the weak perspective camera. This approximation to perspective viewing has been used in many vision systems [250, 56, 201, 287]. Weak perspective is a 512 limiting form of perspective which occurs when the depth of objects along the line of sight is small compared with the viewing distance. This approximation is carried out as follows. Starting with the general perspective matrix, T R2 (R2 O) I3 f (i3 . o) ....

Thompson, D.W. and Mundy, J.L., Three-dimensional Model Matching from an Unconstrained Viewpoint, Proc. ICRA, p.208-220, April 1987.

....on fragmented boundary structures, even if these structures are geometrically accurate. It is clear that accurate geometric information can be reliably extracted from intensity images as demonstratedby the assortment of working vision systems that have appeared over recent years in the literature [1, 2, 14, 15, 25, 31, 39]. These systems generally assume that only fragmentary topological structure is available. For example, two adjacent boundaries meeting at a vertex is considered to be a major feature discovery. By carrying out exhaustive grouping it is possible to align specific 3D models with set of ....

Thompson, D.W. and Mundy, J.L. "Three-Dimensional Model Matching from an Unconstrained Viewpoint," Proceedings ICRA, p.208-220, 1987.

No context found.

D. W. Thompson and J. L. Mundy. Three-dimensional model matching from an unconstrained viewpoint. In Proceedings of the IEEE Conference on Robotics and Automation, pages 208--220, 1987.

No context found.

Thompson D.W. and Mundy J. Three-dimensional model matching from an unconstrained viewpoint. In Proc. IEEE International Conference on Robotics and Automation, pages 208--220, 1987.

No context found.

D. Thompson and J.L. Mundy, "Three-Dimensional Model Matching from an Unconstrained Viewpoint", Proc. IEEE Conf. Robotics and Automation, pp. 208-220, 1987.

No context found.

D. Thompson and J.L. Mundy, "Three-Dimensional Model Matching from an Unconstrained Viewpoint", Proc. IEEE Conf. Robotics and Automation, pp. 208-220, 1987.

No context found.

Thompson DW, Mundy JL (1987) Three-dimensional model matching from an uncon- strained viewpoint. In Proceedings of IEEE Conference on Robotics and Automation, pp 208 220, Raleigh, NC

No context found.

D. W. Thompson and J. L. Mundy. Three-dimensional model matching from an unconstrained viewpoint. In Proceedings of IEEE Conference on Robotics and Automation, pages 208 220, Raleigh, NC, 1987.

No context found.

[Thompson87] Thompson, D. W., and J. L. Mundy, "Three-Dimensional Model Matching from an Unconstrained Viewpoint," in Proc. IEEE Conf. Rob. Aut., pp. 208-220, 1987.

No context found.

Thompson, D.W., Mundy, J.L. (1987), "Three-dimensional model matching from an Unconstrained Viewpoint", IEEE International Conference on Robotics and Automation, pp. 208-220.

First 50 documents  Next 50

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