D.G. Lowe. The viewpoint consistency constraint. Int. J. of Comp. Vision, 1(1):57--72, 1987.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....scenes. Preliminary modeling and recognition results are presented. 1 Introduction This paper addresses the problem of recognizing threedimensional (3D) objects in photographs. Traditional feature based geometric approaches to this problem, for example alignment and interpretation trees [6, 8], enumerate all triples of image features before pose consistency constraints can be used to confirm or discard competing match hypotheses. Appearance based techniques, on the other hand, use rich local descriptions of the image brightness pattern to select a relatively small set of promising ....

....its right and top side. In particular (and not surprisingly) a match between images of the same affine invariant patches contains exactly the same information as a match between triples of points. It is thus clear that all the machinery of structure from motion [3, 5, 23] and pose estimation [6, 8] from point matches can be exploited in our modeling and object recognition tasks. Reasoning in terms of multi view constraints associated with the matrix will provide in this paper a unified and convenient representation for all stages of both tasks, but one should always keep in mind the ....

[Article contains additional citation context not shown here]

D.G. Lowe. The viewpoint consistency constraint. IJCV, 1(1):57--72, 1987.

....scenes. Preliminary modeling and recognition results are presented. 1 Introduction This paper addresses the problem of recognizing threedimensional (3D) objects in photographs. Traditional feature based geometric approaches to this problem, for example alignment and interpretation trees [6, 8], enumerate all triples of image features before pose consistency constraints can be used to confirm or discard competing match hypotheses. Appearance based techniques, on the other hand, use rich local descriptions of the image brightness pattern to select a relatively small set of promising ....

....right and top side. In particular (and not surprisingly) a match between m 2 images of the same affine invariant patches contains exactly the same information as a match between m triples of points. It is thus clear that all the machinery of structure from motion [3, 5, 23] and pose estimation [6, 8] from point matches can be exploited in our modeling and object recognition tasks. Reasoning in terms of multi view constraints associated with the matrix S will provide in this paper a unified and convenient representation for all stages of both tasks, but one should always keep in mind the ....

[Article contains additional citation context not shown here]

D.G. Lowe. The viewpoint consistency constraint. Int. J. of Comp. Vision, 1(1):57--72, 1987.

....speed. We are not aware of any previous attempt to use contour outlines for object localisation as described here, but several authors have made probabilistic arguments based on different feature detection paradigms: 8] used decision theory and likelihood ratios for matching facial features, and [5, 9] both accept hypotheses based on the probability that straight lines and point features should fall in certain configurations. The approach to object localisation presented in this paper applies to objects with complex outlines which need not contain corners or straight lines. It requires no ....

D.G. Lowe. The viewpoint consistency constraint. Int. J. Computer Vision, 1:57--72, 1987.

....projective transformations are collinearity, tangency and incidence conditions, such as intersection and concurrence. This paucity of invariant geometric properties is a major contributor to the difficulty of object description and recognition under perspective viewing. For example, the SCERPO [201] system assumes affine projection in order to use parallelism as 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 ....

....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) ....

Lowe, D.G., The Viewpoint Consistency Constraint, IJCV-1, No. 1, p.57-72, 1987.

....on the model adopted, and will also affect the matching technique selected. Extensive surveys describing methods of object representation and matching have been made by Chin and Dyer [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 ....

D.G.Lowe. The viewpoint consistency constraint. It. Joural of Computer Visions, 1:57 72, 1987.

....of the UMass VISIONS system [10] Another possibility is to use a consistent subset to filter measurement information such that only measurements are considered which support the current hypothesis. This has been done e.g. by D.G. Lowe with his viewpoint consistency constraint in the Scerpo system [13]. Such a consistency constraint can also be an efficient means of outlier rejection. 3.2 Representations A distance measure is required to be able to calculate any measure of consistency. Useful definitions of distance depend on the representation of the problem itself. The following list gives ....

D.G. Lowe. The viewpoint consistency constraint. International Journal of Computer Vision, 1(1):57--72, 1987.

....The full problem is e ectively solved in two steps: a recognition phase followed by a texturing phase. During the recognition phase, a mapping is established between the images and the model scene, this mapping has to be suciently accurate in order to satisfy the viewpoint consistency constraint [12]. Hence, this former problem is twofold: find a general transform so as to give an optimal estimation of the sensor model, reduce the combinatorial explosion of the initial 3D model to image correspondences. The texturing phase allows to decide for the best combination of patches to be ....

D. G. Lowe, The viewpoint consistency constraint, Int. J. Computer Vision, vol. 1, pp. 57-72, 1987.

....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]

David G. Lowe. The Viewpoint Consistency Constraint. International Journal of Computer Vision, 1(1):58 -- 72, 1987.

....similarity we weight the exponential function by this value. Alternatively, one could weight by l Mi . Many approaches for line matching compare the central points or the starting and end points of the model contour with the edge line supposing that the two edges are similar in length (e.g. Lowe [13] , Beveridge et al. 2] In our application the grey value edges of the lower and upper part of the arms and the legs often are connected to one single grey value edge. Then a comparison between central points or starting and end points would lead to large discrepancies. Therefore, we first cut ....

D.G. Lowe, The Viewpoint Consistency Constraint, Intern. J. of Computer Vision 1 (1987) 57-72

....object representation scheme through synthetic examples and image contours detected in real images. 1 Introduction Most approaches to model based object recognition are based on establishing correspondences between viewpoint independent image features and geometric features of object models [9, 15]. For objects with smooth surfaces, few surface markings and little texture, the most reliable image feature is the object s silhouette, i.e. the projection into the image of the curve, called the occluding contour, where the cone formed by the optical rays grazes the surface [11] The ....

D. G. Lowe. The viewpoint consistency constraint. Int. J. Computer Vision, 1(1):57--72, 1987.

....This problem is effectively solved in two steps: a recognition phase followed by an texturing phase. During the recognition phase, a mapping is established between the images and the model scene, this mapping has to be sufficiently accurate in order to satisfy the viewpoint consistency constraint [Lowe, 1987]. Hence, this former problem is twofold: 1. find a general transform so as to give an optimal estimation of the sensor model, 2. reduce the combinatorial explosion of the initial model to image correspondences. The texturing phase permits to vote for the best combination of patches to be ....

Lowe, D. G. (1987). "The viewpoint consistency constraint." Int. J. Computer Vision 1: 57-72.

....plane as we did here. However, he also assumed that there was always an environment map with the landmarks marked on it available. This assumption is valid for calibration, but it may not be valid for other applications such as navigation where there might be no local map available at all. Lowe[9] used viewpoint 2 An analysis of the sensitivity of our algorithm to deviations from this assumption is presented in Section 4. In addition, the algorithm can be extended to situations in which the camera axis is not parallel to the ground plane. 1 consistency constraint to match a set of ....

D. Lowe. The viewpoint consistency constraint. IJCV, 1(1), 1987.

....landmarks available. There is no error analysis for their algorithm. Their algorithm also requires three images, in general, to give the pose information each time. Our algorithm, which is homing based, uses only one image, but it assumes that the target pose information is given in advance. Lowe[11] used viewpoint consistency constraint to match a set of characteristic features against models and achieved relatively good results, although he did not report error measures. For calibration, accuracy is usually the most important criterion, while other applications may have less stringent ....

D. G. Lowe, "The Viewpoint Consistency Constraint", IJCV, Vol.1, No.1, 1987.

....Thus no such false match between scene basis and model basis will be hypothesized. Since we are dealing with highly occluded scenes, we have found that even if a hypothesis has passed verification (by verifying its boundary) it can still be a false alarm. By the viewpoint consistency principle [17], the locations of all object features in an image should be consistent with the projection from a single viewpoint. We may show that all 2 D objects lying on the same plane have an identical ratio of enlargement of area, if viewed from the same camera, assuming approximation of affine ....

D. G. Lowe. The Viewpoint Consistency Constraint. Int. J. of Computer Vision, 1(1):57--72, 1987.

....be then successively applied at larger scales) Lowe and Binford [58] did recognize a role for top down processes, but only in terms of contextual information providing a basis for reduction in the search space. The Witkin and Tenenbaum idea of non accidentalness was adopted and extended by Lowe [56, 57] in developing his SCERPO system for visual recognition. In defining which things are non accidental, Lowe focussed on those relationships between features that are invariant to changes of viewpoint. Thus, he effectively implemented the non accidentalness argument with the constraint that ....

....list they adopted a particular definition of non randomness. tions of lines at a continuous curve (which are all viewpoint invariant) to use as features and relationships for grouping. Given that the problem of organizing object features from an image is severely underconstrained, Lowe [57] subsequently introduced the viewpoint consistency constraint to further reduce the size of the search required. The idea behind this is simply that the positions of all of the elements of an object in an image must be consistent with being viewed from a single point. A simple and obvious point ....

D. G. Lowe. The viewpoint consistency constraint. International Journal of Computer Vision, 1:57--72, 1987.

....hence are used as key features. A variety of key feature algorithms have used local curvature properties to identify object silhouettes [KJ86, GTM89, AD90] This silhouette recognition problem is simpler than recognition in grey scale images since the input is assumed to be a binary image. Lowe [Low85, Low87] has made two major contributions [Low85] to the key feature approach: 1) he argued that human perceptual organization performs feature grouping in order to find key features which aid in recognition, and 2) he demonstrated with a working algorithm that key features can play a significant role in ....

....and 2) he demonstrated with a working algorithm that key features can play a significant role in recognizing 3D objects. Lowe also did a good job of quantitatively handling the problems associated with registering a 3D wire frame model to corresponding segments in an image. His early work [Low85, Low87] assumed scaled orthographic projection while his subsequent work on object tracking [Low91] handles 3D perspective. More recent work with 3D models and an approach somewhat like the key feature approach is that of Huttenlocher [HU90] In the algorithm developed by Huttenlocher, feature triples ....

David G. Lowe. The viewpoint consistency constraint. Iternational Journal of Computer Vision, 1(1):58 -- 72, 1987.

....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 sequence of frames where the previous frame gives a good guess for the current frame, as demonstrated in [24], or in human computer interfaces where a good initial guess can be provided by a user, as in [27] However the problems with gradient descent in non convex spaces is well documented, and the major problem with these methods is their reliance on prior estimates combined with inability to ....

D. G. Lowe. The viewpoint consistency constraint. International Journal of Computer Vision, 1(1):57--72, 1987.

....aligns a model to image features [31] to yield an initial estimate of pose. This hypothesised alignment is tested by searching for other model to image correspondence predicted by the model pose (verification) This algorithm has been implemented for a variety of data formats and feature types [1, 6, 15, 24, 38, 69]. In fact, extensions to 3D curved surfaces have even been created [13, 33] 3. Pose clustering is implemented by computing the object pose from a group of features corresponding to a particular model, and storing the estimate in an accumulator in pose space; if enough local groups have the same ....

....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 reasonable simply to try to find image feature support for each model. This approach is typical of many existing systems [1, 2, 28, 31, 38, 48, 52]. As the size of the model library increases, this approach becomes computationally too expensive. It is then more effective to choose potential models from the library based on the observed image features. That is, image feature measurements are used to index into the model base. In constructing ....

Lowe, D.G. "The Viewpoint Consistency Constraint," International Journal of Computer Vision, Vol. 1, No. 1, p.57-72, 1987.

....to model feature assignments in order to form hypotheses. Such systems fall into two main categories: interpretation tree techniques (such as those of Ayache and Faugeras [1] or Grimson and Lozano P erez [6] or transformation determination methods (those of Huttenlocher and Ullman [7] Lowe [10] or Mundy and Heller [11] For an object recognition system with a large number of models, testing for the presence of each model becomes impractical. Instead, it is necessary to introduce the concept of indexing functions. These provide direct access to a certain model in the data base without ....

Lowe, D.G. "The Viewpoint Consistency Constraint," IJCV-1, No. 1, p.57-72, 1987.

No context found.

D.G. Lowe. The viewpoint consistency constraint. Int. J. of Comp. Vision, 1(1):57--72, 1987.

No context found.

D.G. Lowe, "The Viewpoint Consistency Con- straint," International Journal of Computer Vision, vol. 1, no. 1, 1987, pp. 57-72.

No context found.

D. G. Lowe, \The Viewpoint Consistency Constraint," International Journal of Computer Vision, Vol. 1, 1987, pp. 57072.

No context found.

D. G. Lowe, \The Viewpoint Consistency Constraint," International Journal of Computer Vision, Vol. 1, 1987, pp. 57072.

No context found.

Lowe, David G. (1987) "The Viewpoint Consistency Constraint," Intern. J. of Comp. Vis, 1/1, pp. 57--72.

No context found.

Lowe, David G. (1987) "The Viewpoint Consistency Constraint," Intern. J. of Comp. Vis, 1/1, pp. 57--72.

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