G. Stockman. Object recognition and localization via pose clustering. Computer Vision, Graphics, and Image Processing, 40:361--387, 1987.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....speeds, since the tree search is exponential [3] This is due to the fact that an interpretation tree is forming the O(m matches between model and sensed features. Incorporating these further techniques, the outline of Grimson s approach is as follows. Perform an initial pose clustering (as in [4]) Perform multiple interpretation tree searches on the reduced sets of matched features revealed through the clustering. Employ a mismatch tolerance threshold to terminate search to allow for sensor noise and occlusion. The pose clustering method described by Stockman [4] does not reveal the ....

....pose clustering (as in [4] Perform multiple interpretation tree searches on the reduced sets of matched features revealed through the clustering. Employ a mismatch tolerance threshold to terminate search to allow for sensor noise and occlusion. The pose clustering method described by Stockman [4] does not reveal the additional combinatorial increase of the number of high level sensed and model structures that can be formed from lower level iconic features, e.g. line segments and constant curvature arcs. For example, for m model features, we must form (m 1) structures from these ....

G. Stockman, "Object recognition and localization via pose clustering, " in Computer Vision: Advances and Applications, Rangachar Kasturi and Ramesh C. Jain, Eds. 1991, IEEE Computer Society Press.

....unique determination of the object s pose. Initial Guess for Object s Pose All emerged segment triplets are searched, primarily to build a correspondence tree for each position cluster without pose restrictions. Neighborhood relationships such as the angle and distance between two segment lines [6] are criteria for establishing correspondences between image and reference segments. Matching candidates for the second and third feature of a group must be searched only in certain direction clusters, reducing the number of required comparisons. A prehypothesis is established if all three ....

G. Stockman. Object recognition and localization via pose clustering. Computer Vision, 40:361--387, 1987.

....assignment (a mapping seed) It then adds to this mapping, employing a consistency check as nodes are added to the subgraph. A number of common methods for object recognition combine the problems of recognition and pose determination. Methods such as local feature focus [3] pose clustering [24], and geometric hashing [25] for example. These methods are able to employ additional constraints, generally involving the physical location of graph nodes, to limit potential matches. Some methods of image registration also use this type of approach [7] While these application areas are of ....

G. Stockman, Object recognition and localization via pose clustering, Computer Vision, Graphics and Image Processing, 40 (1987) 361-387.

....created by the constraint sets[1] implied by correspondences between model features and image features. Directly applied, such methods do not appear to be practical. However, the insights they are based on form the basis for the algorithm presented in this paper. Pose clustering techniques[13] are based on examining the transformations ( poses ) implied by many di#erent hypothesized correspondences between image and model features. Transformations that bring many image and model points into correspondence under given error bounds will tend to cluster in the space of transformations. ....

Stockman, G., Object recognition and localization via pose clustering, Computer Vision, Graphics, and Image Processing vol.40, no.3 (1987), pp. 361--87. 14

....vision and pattern recognition. Some works (e.g. 14] depend on the ability to match significant features of the objects, like knobs and holes, whose existence is not usually guaranteed. Other methods, which do not rely on the existence of a particular type of features, are pose clustering [78], alignment [49] and geometric hashing. A comparison between these techniques is found in [83] Comprehensive surveys on partial surface matching techniques in computer vision are found in [11] 23] Many other works have addressed the problem, most of which have various limitations. They either ....

G. Stockman, "Object Recognition and Localization Via Pose Clustering," Computer Vision, Graphics, and Image Processing, vol. 40, pp. 361-387, 1987.

....of exact, provably correct and efficient solutions to geometric problems. First some related work is mentioned in the next subsection. 1. 1 Related work Matching has been approached in a number of ways, including tree pruning [55] the generalized Hough transform [8] or pose clustering [51], geometric hashing [59] the alignment method [27] statistics [40] deformable templates [50] relaxation labeling [44] Fourier descriptors [35] wavelet transform [31] curvature scale space [36] and neural networks [21] The following subsections treat a few methods in more detail. They are ....

....between the two shapes. The complexity of matching a single query set of # points is ####. There are several variations of this basic method, such as balancing the hashing table, or avoiding taking all possible ### # ## tuples. The generalized Hough transform [8] or pose clustering [51], is also a voting scheme. Here, affine transformations are represented by six coefficients. The quantized transformation space is represented as a six dimensional table. Now for each triplet of points in one set, and each triplet of points from the other set, compute the transformation between ....

G. Stockman. Object recognition and localization via pose clustering. Computer Vision, Graphics, and Image Processing, 40(3):361--387, 1987.

....[11] for an introduction and [19] and [20] for a survey) work by computing the set of transformations determined by all correspondences between pairs of image points and pairs of model points. Clusters of these transformations are then found using a binning scheme in transformation space (see also [25] for a discussion of this view) Because of the presence of error, the transformations resulting from the optimal solution to the bounded error matching problem form a cloud in transformation space, not just a single point. With the Hough transform, this cloud may get split up among several bins ....

G. Stockman. Object recognition and localization via pose clustering. Computer Vision, Graphics, and Image Processing, vol.40, no.3:361--87, 1987.

....Therefore the approximate value for Z has to be good enough for the matching step. To get a more scale independent orientation procedure it may be suitable to change the pose clustering step. Here not only a translation, but also a scale and a rotation parameter could be estimated, as described in (Stockman 1987). 2 Work flow used for the test 2.1 New Problems AMOR had not been tested on large scale aerial photographs and new problems to be solved arose. The following table gives an overview. 3 Problem Solution search area too large in original image using other pyramid levels buildings are too ....

Stockman, G. (1987). Object recognition and localization via pose clustering. Computer Vision, Graphics and Image Processing 40, 361--387.

....[7, 11] and in a companion paper [4] Some works (e.g. 9] depend on the ability to match significant features of the objects, like knobs and holes, whose existence is not usually guaranteed. Other methods, which do not rely on the existence of a certain type of features, are pose clustering [30], alignment [19] and, of course, geometric hashing. A comparison between these techniques is found in [31] Many other works have addressed the problem; see [3, 5, 6, 8, 10, 13, 17, 18, 27] for studies in the context of object recognition, and [1, 12, 14, 20, 22, 24, 26, 28] for studies in the ....

G. Stockman, Object recognition and localization via pose clustering, Computer Vision, Graphics, and Image Processing, 40 (1987), 361--387.

....vision and pattern recognition. Some works (e.g. 14] depend on the ability to match significant features of the objects, like knobs and holes, whose existence is not usually guaranteed. Other methods, which do not rely on the existence of a particular type of features, are pose clustering [78], alignment [49] and geometric hashing. A comparison between these techniques is found in [83] Comprehensive surveys on partial surface matching techniques in computer vision are found in [11] 23] Many other works have addressed the problem, most of which have various limitations. They either ....

G. Stockman, "Object Recognition and Localization Via Pose Clustering," Computer Vision, Graphics, and Image Processing, vol. 40, pp. 361-387, 1987.

....obtenir le point d accumulation central correspondant a une hypothese interessante pour le recalage. RR n 2716 10 J. P. Tarel, N. Boujemaa d utiliser une methode de classification est de ne pas e tre penalise par la discretisation de l espace d accumulation comme dans la transformee de Hough [Sto87] 2.2.1 L approche floue La modelisation des problemes de vision par ordinateur avec des sous ensembles flous [Bou93] permet de reporter les prises de decision en phase ultime d un processus de traitement, minimisant ainsi la propagation des erreurs decisionnelles dans des phases ....

Stockman (G.). -- Object recognition and localization via pose clustering. Computer Vision, Graphics and Image Processing, vol. 40, n 2, 1987, pp. 361--387.

....tree pruning methods [1, 3] make hypotheses concerning the correspondence by searching over a tree in which each node represents a partial match. Each partial match is then evaluated through the pose that best fits it. In the generalized Hough transform or equivalently template matching approach [7, 3], optimal transformation parameters are computed for each possible pairing of a model feature and a scene feature, and these optimal parameters then vote for the closest candidate in the discretized transformation space. By contrast with the tree pruning methods and the generalized Hough ....

G. Stockman. Object recognition and localization via pose clustering. Computer Vision, Graphics, and Image Processing, (40), 1987.

....and strike the image plane. A common strategy for object recognition and pose estimation is to match individual features in the image to features on the model [27] One category of methods employ schemes such as the Hough transform to search pose space in order to identify objects in images [28]. Another category employs treebased search schemes to search a correspondence space [29] A depth first approach is used to search interpretation trees for matches between object and image, where each node of the tree represents a pairing of feature of the object to a feature of the image. ....

G. Stockman, "Object recognition and Localization Via Pose Clustering," Comput Vision Graphics Image Process, Vol. 40, No. 3, pp. 361-387, 1987.

....measure. Figure 2: Fingerprint matching. Figure 3: Query hieroglyph (left) and hieroglyphs 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 ....

....the query and the target. The complexity of matching a single query set of n points is O(n) There are several variations of this basic method, such as balancing the hashing table, or avoiding taking all possible O(n 3 ) 3 tuples. The generalized Hough transform [Bal81] or pose clustering [Sto87] is also a voting scheme. Here, a ne transformations are represented by six coe cients. The quantized transformation space is represented as a six dimensional table. Now for each triplet of points in the query set, and each triplet of points from the target set, compute the transformation ....

G. Stockman. Object recognition and localization via pose clustering. Computer Vision, Graphics, and Image Processing, 40(3):361-387, 1987.

....model features [8, 9, 10, 11, 12, 13] 3. Vote Accumulation: Hypotheses are generated by selecting scene feature subsets that cover all scene features, and aligning them with model subsets. These hypotheses are then clustered, and the one corresponding to the cluster of largest size is selected [14, 15, 16]. Choice of the appropriate approach for a given recognition task depends on several factors, such as amount of sensor noise, scene complexity (number and configuration of scene objects) and dimensionality of scene data (2 D or 3 D) Object recognition systems can also be classified into two ....

G. Stockman. Object recognition and localization via pose clustering. Comput. Vision Graphics Image Process., 40(3):361--387, 1987.

....algorithms search for highly distinctive local geometric features which indicate the placement of an object in a scene; the local feature focus method of Bolles and Cain [BC82] is an early example. Generalized Hough: Generalized Hough algorithms [DY80, Bal81] and more recently pose clustering [Sto87] algorithms, emphasize search in the space of object model to image pose transformations. Tree Search: Tree search algorithms [Bai85, GLP87, Gri90c, Bre90, Cas92] expand a tree of potential matches between model and image features using local geometric consistency constraints for pruning. ....

....are used to predict object identity as well as placement. Key feature matching [BC82, Low85] covers a very broad range of specific algorithms, and is one of the more intuitive approaches. The Generalized Hough approach [DY80, Bal81] emerged at essentially the same time, although pose clustering [Sto87] was formulated later. The early contributions in tree search [Bai85, GLP87, Gri90c] came later than either key feature or generalized Hough. Exciting extensions to tree search have recently appeared [Bre90, Cas92] Finally, geometric hashing [KSSS86, LW88] is the most recent approach to be ....

[Article contains additional citation context not shown here]

George Stockman. Object recognition and localization via pose clustering. Computer Vision, Graphics, and Image Processing, 40:361 -- 387, 1987.

....on surface and volume matching. Some works (e.g. BH] depend on the ability to match significant features of the objects, like knobs and holes, whose existence is not usually guaranteed. Other methods, which do not rely on the existence of a certain type of features, are poseclustering [St], alignment [HU] and, of course, geometric hashing. A comparison between these techniques is found in [Wo1] Comprehensive surveys on partial surface matching techniques in computer vision are found in [BJ, CD] Relatively little work has been published in the area of registration (pose ....

G. Stockman, Object recognition and localization via pose clustering, Computer Vision, Graphics, and Image Processing, 40 (1987), 361--387.

.... A preliminary version of this work appears in (Olson, 1994) from hypothesized matches between feature sets in the object model and feature sets in the image (Ballard, 1981; Stockman et al. 1982; Silberberg et al. 1984; Turney et al. 1985; Silberberg et al. 1986; Dhome and Kasvand, 1987; Stockman, 1987; Thompson and Mundy, 1987; Linnainmaa et al. 1988) In this method, the transformation parameters that bring the sets of features into alignment are determined. Under a rigidbody assumption, the correct matches will yield transformations close to the correct pose of the object. Objects can thus ....

Stockman, G. 1987. Object recognition and localization via pose clustering. Computer Vision, Graphics, and Image Processing, 40:361--387.

.... i ; P i ) p j ; P j ) p k ; P k ) is consistent if kp i Gamma p j k kP i Gamma P j k = kp j Gamma p k k kP j Gamma P k k = kp k Gamma p i k kP k Gamma P i k (33) The above equations test the conditions for the two triangles (p i ; p j ; p k ) and (P i ; P j ; P k ) to be similar [35]. To deal with inaccuracy and noise, lower and upper bounds (0.8 and 1 0.8 in our implementation) are used in testing the above conditions. The test is done for all 1 i N Gamma 2, i 1 j N Gamma 1 and j 1 k N , yielding a set of pose candidates. Best pose candidates are selected using a ....

G. Stockman. "Object recognition and localization via pose clustering". Computer Vision, Graphics and Image Processing, 40:361--387, 1987.

....are consequently expensive to detect. Sitarman and Rosenfeld [SR89] have theorized that there is some optimum size for key features, but how best to select and detect key features remains an open question. Pose Space Search Generalized Hough transform [DY80, Bal81, Dav82] and pose clustering [Sto87] algorithms shift search from correspondence space and into pose space. Illingworth and Kittler [IK88] review generalized Hough work through 1988. Explicit representations of pose space place practical limits on the dimensionality of the pose space and most algorithms match 2D rigid objects. ....

....through 1988. Explicit representations of pose space place practical limits on the dimensionality of the pose space and most algorithms match 2D rigid objects. Notable exceptions include Silberberg [SHD86] who limited camera placement and Thompson [TM87] who used subspace projection. Stockman s [Sto87] pose clustering avoids explicit representations of pose space and a nice illustration of pose clustering can be found in the work of Hwang [Hwa89] The noise sensitivity of generalized Hough algorithms has been assessed by Brown [Bro82] and Grimson [GH88] Grimson concludes that false positive ....

[Article contains additional citation context not shown here]

George Stockman. Object recognition and localization via pose clustering. Computer Vision, Graphics, and Image Processing, 40:361 -- 387, 1987.

....Hough) has the disadvantage that the pose space is high dimensional (six dof for 3D Euclidean space) so searching for consistent pose is expensive. Two ways round this are to use a decomposition of the pose space into separable parameters [44, 68] or to use an adaptive Hough transform [66]. Another approach eliminates the requirement to quantise the pose space into rectangular cells by constructing a quantisation that depends both on the estimates of pose, and on the expected error bounds of the pose measurements [10] For a small number of models, for example two or three, it is ....

Stockman, G. "Object Recognition and Localization via Pose Clustering," Computer Vision Graphics and Image Processing, Vol. 40, p.361-387, 1987.

....2 Recognition using image based aspects There has been a great deal of CV work within the recogniton by alignment paradigm. Often, alignment is achieved by the ability to extract salient features from the observed data and make correspondences between those features and model features (e.g. [20, 14, 11]) Only one global model containing the aggregation of labeled geometric features is needed. Another significant advantage of this approach is that an alignment transformation can be computed in closed form from a small set of feature correspondences. Disadvantages are that (a) a general method ....

G. Stockman, Object Recognition and Localization via Pose Clustering, CVGIP, Vol. 40, 1987, pp.361-387.

No context found.

G. Stockman. Object recognition and localization via pose clustering. Computer Vision, Graphics, and Image Processing, 40:361--387, 1987.

No context found.

G.Stockman, "Object recognition and localization via pose clustering", Computer vision, Graphics, and Image Processing, 40(3):361-387,1987.

No context found.

Stockman, G., Object Recognition and Localization via Pose Clustering, CVGIP-40, No. 3, p.361-387, 1987.

First 50 documents

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