D. P. Huttenlocher and S. Ullman. Recognizing solid objects by alignment. Int. J. of Computer Vision, 5(2):195--212, 1990.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....Knowing that a given subset of points of an object is projected on the image into a given subset of points determines constraints on the object location in space. The validity of a match is estimated either by verifying that the object at the found location does project onto the given image [7] or by checking whether many other matches determine a similar location. Fisohler and Bolles [4] coined the term Perspective n point problem for the problem of finding the position and orientation of an object from the images of n points at known locations on the object. Their paper provides ....

....insights to this problem (sec also [6] for a review of solutions and a solution for four points) The question of how many points should be taken as subsets in an object pose or recognition system has generated some interest. Many researchers have considered three point solutions [10] 12] 9] [7] because it is the smallest subset that yields a finite number of object poses, generally two poses (four poses in some image configurations) The perspective three point problem, which is also called the triangle pose problem [10] has been solved in many different ways. A review of the major ....

[Article contains additional citation context not shown here]

D. Huttenlocher and S. Ullman, "Recognizing solid objects by align- ment," in Proc. 1988 DARPA Image Understanding Worhop, pp. 1114-1122.

.... Examples include image to image registration for mosaic construction [23, 40, 37] range data to range data registration for model construction in reverse engineering [3, 19, 29, 35, 9] and model to image registration for tracking, motion estimation, object recognition or camera calibration [14, 22, 27, 28, 44]. The particular instance of the registration problem considered here is registering a three dimensional model to one or more range data sets as precisely and accurately as possible. This problem, which di ers from the range data to range data registration problem because the model is known in ....

D. P. Huttenlocher and S. Ullman. Recognizing solid objects by alignment. Int. J. of Computer Vision, 5(2):195-212, 1990.

....are properly obtained (e.g. camera is kept horizontal or at a xed tilt angle as it rotates about the vertical to obtain the panoramic image) 2. POSEESTIMATIONUSINGALIGNMENT Our proposed technique is based on the alignment principle which consists of a two step hypothesize verify process [10]. In the rst stage, image plane alignmentofagroupof model (DEM) features with a group of image features is hypothesized. If it exists, the transformation or pose that maps the model features onto their corresponding image features is then computed. In the second stage, the transformation is ....

D. Huttenlocher and S. Ullman, \Recognizing Solid Objects by Alignment," Pr oc. DARPAImage Understanding Workshop, pp. 1114-1124, 1988.

....a perceptual grouping relation. Also, affine geometry is often assumed in model based vision 2A call taken up by Naeve and Eklundh [216] 466 The concept of parallelism is not meaningful for perspective projection. Notice that parallel lines converge to a vanishing point at the horizon. systems [56, 159] because fewer features are required to compute model pose uniquely under arline projection than perspective projection. However, the arline approximation to perspective fails when the depth range of an object is significant compared to the viewing distance. The most important aspect of ....

Huttenlocher, D.P. and Ullman, S., Recognizing Solid Objects by Alignment, IJCV-5, No. 2, p.255-274, 1990.

....algorithm because it is not a phenomena which can be predicted in isolation: occlusion is a function of an object s relationship to the scene in which it is embedded. Traditional recognition techniques either rely on static feature measurements remaining stable in the presence of occlusions [19, 12, 3], or associate a likelihood of finding each feature based on off line appearance analysis [6, 26, 1, 11] In most of these works, some occlusion is tolerated, but it is seldom dealt with explicitly. Instead, a match quality metric ranks potential matches, and matches with missing features are ....

....paying particular attention to how occluded objects are handled. 2.1. Geometric Feature Matching Geometric object recognition centers around the search for correspondences between geometric model features, such as points, lines, planes, etc. and homogeneous features extracted from sensor data [19, 12, 3, 25]. While a variety of different methods have been developed, most require the construction of a correspondence set in order to form a match. Such a set contains tuples of model features matched to one or more data features. To be considered valid, these matches must remain topologically consistent ....

[Article contains additional citation context not shown here]

Daniel P. Huttenlocher and Shimon Ullman. Recognizing Solid Objects by Alignment. In Proc. of the DARPA Image Understanding Workshop, pages 1114 -- 1124, Cambridge, April 1988. Morgan Kaufmann.

....the objective function components are clearly related to specific probabilistic models. A second advantage is that the trade off parameters in the objective function can be derived from measurable statistics of the domain. 1. 2 Alignment The basic strategy of the alignment method of recognition [6] is to use separate mechanisms for generating and testing hypotheses. Recently, indexing methods have become available for efficiently generating hypotheses in recognition. These methods avoid a significant amount of search by looking up, in precomputed tables, the object features that might ....

D.P. Huttenlocher and S. Ullman. Recognizing Solid Objects by Alignment. In Proceedings: Image Understanding Workshop, pages 1114--1124. Morgan Kaufmann, April 1988.

.... they revealed that all the Perspective Three Point(P3P) problems can have as many as four possible solutions (excluding the reflection) Formation of these multiple solutions was later explained by Wolfe, Mathis, Sklair and Magee (Wolfe et al. 1991) On the other hand, Huttenlocher and Ullman (Huttenlocker and Ullman, 1988) proved that 3 non collinear model points are enough to align a model, however, it is under weak perspective projection. Our algorithm is also similar to that in (Lowe, 1987) and (McReynolds and Lowe, 1996) but we use the length of the vectors from the focus point of the camera to the features as ....

Huttenlocker, D. P. and Ullman, S. (1988). Recognizing solid objects by alignment. presented at the DARPA Image Understanding Workshop, Cambridge MA, pages 1114--1122.

.... condition estimation uses geometric cues such as 3 D object shape models and geometric relationships between object features, and recent appearance based approaches describe an object by using relatively compact eigenspaces derived only from image appearances under various viewing conditions [5] [17] [33] 25] Although color reflectance has been conceived as an obvious object descriptor and color matching has been a central focus of color science and engineering, only recently have a number of researchers begun to explore the use of color distributions as signatures for object recognition. ....

D.P. Huttenlocher and S. Ullman. Recognizing solid objects by alignment. IJCV, 5, 1990.

....algorithms used depends largely on the limited scope of the representation; objects are modelled as either a box with three scaling degrees of freedom or cylinders with two scaling degrees of freedom. The major research on 3D parametric object recognition from 2D data has been conducted by Lowe [23] (an extension of the SCERPO system) and Nguyen et al. 27] Both methods use gradient descent from an a priori estimate of parameter values (including pose parameters) to solve simultaneously for pose and parameters. Such systems are extremely useful for tracking articulated objects from a dense ....

D. P. Huttenlocher and S. Ullman. Recognizing solid objects by alignment. International Journal of Computer Vision, 5(2):255--274, 1990.

.... to assuming that the involved image points have been obtained by a scaled orthographic projection (SOP in the following) We refer to this part of the algorithm as POS (Pose from Orthography and Scaling) The works of Tomasi [26] and Ullman and Basri [28] apply related techniques (see also [17] for related work with three points) The next iterations apply exactly the same calculations, but with corrected image points. The basic idea is that since the POS algorithm requires an SOP image instead of a perspective image to produce an accurate pose, we have to compute SOP image points, ....

Huttenlocher and D., S. Ullman, "Recognizing Solid Objects by Alignment", Proc. DARPA Image Understanding Workshop, pp. 1114--1122., 1988.

....solution method. In the following we will review some, but not all, different methods used for pose determination. 3.2.1 2D 3D Matching 2D features to 3D models is a very common case in the literature. Of the non iterative solutions can be mentioned the work by Huttenlocher and Ullman [HU88]. They showed that given 3 point to point matches, under a weak perspective projection (orthographic projection scale) there exists an analytic solution. Using perspective projection Dh me et al. DRLR88] have presented a method to solve the pose problem using 3 line correspondences. The ....

Daniel P. Huttenlocher and Shimon Ullman. Recognizing solid objects by alignment. In Proceedings of the Image Understanding Workshop, pages 1114--1124. DARPA, April 1988.

....well. A second advantage is that the trade off parameters in the objective function can be derived from measurable statistics of the domain. 1.4. Alignment and Indexing Methods Hypothesize and test, or alignment methods have proven useful in visual object recognition. Huttenlocher and Ullman (Huttenlocher and Ullman, 1988) used search over minimal sets of corresponding features to establish candidate hypotheses. In their work these hypotheses, or alignments, are tested by projecting the object model into the image using the pose (position and orientation) implied by the hypothesis, and then by performing a detailed ....

Huttenlocher, D. and Ullman, S. (1988). Recognizing Solid Objects by Alignment. In Proceedings: Image Understanding Workshop, pages 1114--1124. Morgan Kaufmann.

....Knowing that a given subset of points of an object is projected on the image into a given subset of points determines constraints on the object location in space. The validity of a match is estimated either by verifying that the object at the found location does project onto the given image [7], or by checking whether many other matches determine a similar location. Fischler and Bolles [4] coined the term Perspective n Point Problem for the problem of finding the position and orientation of an object from the images of n points at known locations on the object. Their paper provides ....

....and useful insights to this problem (see also [6] for a review of solutions and a solution for four points) The question of how many points should be taken as subsets in an object pose or recognition system has generated some interest. Many researchers have considered three points solutions [10, 12, 16, 7] because it is the smallest subset which yields a finite number of object poses, generally two poses (four poses in some image configurations) The perspective 3 point problem, also called the triangle pose problem [10] has been solved in many different ways. A review of the major direct ....

[Article contains additional citation context not shown here]

D. Huttenlocher and S. Ullman, "Recognizing Solid Objects by Alignment", Proc. 1988 DARPA Image Understanding Workshop, 1114--1122.

....attainable with other existing vision systems 1 . 1 Introduction Much of the progress that has been made to date in machine vision has been based, almost exclusively, on shape comparison and classification employing locally measurable attributes of the imaged objects (e.g. color and texture) [2, 5, 6, 11, 13]. Natural objects viewed 1 Supported by the Defense Advanced Research Projects Agency under contracts MDA903 86 C 0084, 89F737300, and DACA76 90 C 0021. under realistic conditions do not have uniform shapes which can be matched against stored prototypes, and their local surface properties are ....

....1. All objects of interest are defined by a relatively small number of explicit shape models. This makes it computationally feasible to exhaustively search for the presence of these models (via geometric alignment ) as a way of producing a suitable description of some given scene (as in [4] and [13], for example) 2. All objects of interest have characteristic features, homogeneous and locally measurable in an image (e.g. color or texture) which are reliable indicators of the object s identity. This either allows direct determination of the presence of objects using statistical decision ....

Huttenlocher, Daniel P., and Ulman, Shimon, "Recognizing Solid Objects by Alignment," Proceedings: DARPA Image Understanding Workshop, Cambridge, Massachusetts, April 1988, pp. 1114--1122.

....denote pre synaptic neuronal responses (or firing rates) 2.1 Previous Approaches There has been recent interest in appearance based approaches to computational vision. These differ significantly from traditional 3D model based or geometry based approaches [ Lowe, 1987; Lamdan and Wolfson, 1988; Huttenlocher and Ullman, 1987; Grimson, 1990 ] which have typically been limited to representing restricted types of geometric objects. In the appearance based approach, the need for explicit 3D geometric models of objects is avoided by extracting object representations directly from the input images. For example, Buhmann ....

D. P. Huttenlocher and S. Ullman. Recognizing solid objects by alignment. In International Conf. on Computer Vision, 1987.

....(excluding reflections) They also described specific geometric configurations in which four triangles of the same shape project to a single image triangle. Wolfe et al. 15] later showed that for most of the time there would be only two solutions. On the other hand, Huttenlocher and Ullman [8] proved that three noncollinear image points are enough to align a model under weak perspective projection, and Alter [1] introduced a solution for 3D pose estimation from three points using this projection model. However, the weak perspective assumption is valid only when the distance between the ....

Huttenlocher D.P. and Ullman S., Recognizing Solid Objects by Alignment, Image Understanding Workshop, vol. 2, pp. 1114-1122, Apr. 1988.

....results on the efficacy of the affine approximation to the perspective projection. Section 8 discusses the techniques and results and includes an analysis of the speedup produced under various conditions. Finally, section 9 presents our conclusions. 2 The Alignment Method The alignment method [Huttenlocher and Ullman, 1987, 1988, 1990] is a model based technique for recognizing rigid three dimensional objects from a single two dimensional image. The premise of the alignment method is that a unique (up to a reflection) affine transformation between the model and image of the model can be found by matching three model ....

D. P. Huttenlocher and S. Ullman. Recognizing solid objects by alignment. In Proceedings of the DARPA Image Understanding Workshop, pages 1114--1124, 1988.

....on the efficacy of the affine approximation to the perspective projection. Section 5.8 discusses the techniques and results and gives an analysis of the speedup produced under various conditions. Finally, Section 5.9 summarizes the results. 78 5. 2 The alignment method The alignment method [Huttenlocher and Ullman, 1987, 1988, 1990] is a modelbased technique for recognizing rigid three dimensional objects from a monocular two dimensional image. The premise of the alignment method is that a unique (up to a reflection) affine transformation between the model and image of the model can be found by matching three model ....

D. P. Huttenlocher and S. Ullman. Recognizing solid objects by alignment. In Proceedings of the DARPA Image Understanding Workshop, pages 1114--1124, 1988.

.... they revealed that all the Perspective Three Point(P3P) problems can have as many as four possible solutions (excluding the reflection) Formation of these multiple solutions was later explained by Wolfe, Mathis, Sklair and Magee (Wolfe et al. 1991) On the other hand, Huttenlocher and Ullman (Huttenlocker and Ullman, 1988) proved that 3 non collinear model points are enough to align a model, however, it is under weak perspective projection. Our algorithm is also similar to that in (Lowe, 1987) and (McReynolds and Lowe, 1996) but we use the length of the vectors from the focus point of the camera to the features as ....

Huttenlocker, D. P. and Ullman, S. (1988). Recognizing solid objects by alignment.

....such as points or edges, and their geometric relations are encoded using a minimal set of such features. The task becomes more complex if the models overlap in the scene and or other occluded unfamiliar models exist in the scene. Sequential recognition techniques proposed in the literature [1, 2, 3, 4, 5, 6] differ based on the features used to represent a given model and the searching techniques employed to recognize a scene, and hence vary in terms of robustness. Most of these techniques take several minutes of processing time on a serial machine to execute a single probe of the recognition phase. ....

D. P. Huttenlocher and S. Ullman, "Recognizing solid objects by alignment," DARPA Image Understanding Workshop, 1988, pp. 1114--1124.

No context found.

D. P. Huttenlocher and S. Ullman. Recognizing solid objects by alignment. Int. J. of Computer Vision, 5(2):195--212, 1990.

No context found.

D.P. Huttenlocher and S. Ullman, "Recognizing Solid Objects by Alignment," Proc. of the DARPA Image Understanding Workshop, Vol. II, pp. 1114-1122, Cambridge, MA, April, 1988.

No context found.

D.P. Huttenlocher and S. Ullman, "Recognizing Solid Objects by Alignment," Proc. of the DARPA Image Understanding Workshop, Vol. II, pp. 1114-1122, Cambridge, MA, April, 1988.

No context found.

D. Huttenlocher and S. Ullman (1988), Recognizing Solid Objects by Alignment, Proc. DARPA Spring Meeting1114-1122.

No context found.

D. Huttenlocher and S. Ullman (1988), Recognizing Solid Objects by Alignment, Proc. DARPA Spring Meeting1114-1122.

No context found.

Daniel P. Huttenlocher and Shimon Ullman. Recognizing Solid Objects by Alignment. In Proc. of the DARPA Image Understanding Workshop, pages 1114 -- 1124, Cambridge, April 1988. Morgan Kaufman Publishers, Inc., New York.

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