I.D. Reid and D.W. Murray. Tracking foveated corner clusters using affine structure. In Proceedings of the 4th IEEE International Conference on Computer Vision, Berlin, Germany, 1993, pages 76--83, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....for size constancy, we can approach the problem of zooming for scale change. Changing scale poses increased demands on target identification and localization because commonly used approaches such as cross correlation are not scale invariant. Alternatives include feature based tracking (e.g. [17]) or the use of adaptive correlation templates (e.g. 12] 2.4 National Institute of Standards and Technology (Martin Herman, David Coombs, Sandor Szabo, Tsai Hong Hong) NIST was responsible for developing the vision processing platform, assisting in integrating University software onto the ....

I.D. Reid and D.W. Murray. Tracking foveated corner clusters using affine structure. In Proc. of International Conference on Computer Vision, Berlin, Germany, May 1993, pp. 76--83.

....features or corners on objects. These well located points could be matched between successive frames of a tracking sequence to reveal how certain parts of the object were moving. In order to determine a fixation point to use to track the whole object the method of affine transfer was developed [76], since it was discovered that just using one of the points or a mean of all of those found in a certain frame was unstable (a slightly different set of points would be visible in each frame) This is a way of calculating a stable fixation point on an object from whatever corners are detected in ....

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. 4th Int'l Conf. on Computer Vision, Berlin, pages 76--83, Los Alamitos, CA, 1993. IEEE Computer Society Press.

....of other objects. These features are relatively scarce in the image, furthermore they constrain the detected image point to a ray through the imaged scene, making them useful for applications that require matching points in different images, such as stereo matching [24] and structure from motion [56]. In this thesis, the term 1D image feature refers to image points that have a significant locally maximal variation in at least one orientation. This variation may be a step, roof, or spike discontinuity, or a combination of these events. Note that this definition includes higher order ....

....data set preserves a significant portion of the information in the original image. 2D image features allow this data reduction while constraining the detected image point to a ray through the viewed scene, making them useful for applications such as stereo matching [24] structure from motion [56] and line labeling [17, 27, 8, 73, 35] 2.1 What are 2D Features Early attempts at 2D feature detection were often methods for the detection of greylevel corners (or L junctions) in images [28, 47] These image points typically arise from the projection of points that are object vertices or ....

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In 4th International Conference on Computer Vision, 1993.

....for obtaining a prolonged view of the moving object and for the estimation of the 3D structure and motion parameters. Corners (feature points) are easily computed, and are small in number compared to the overall image size, making corner tracking algorithms feasible for real time implementations [11]. However, the primary problem in point tracking arises due to the localization and detection errors of most corner detectors [16, 6] making it necessary to design tracking algorithms which are insensitive to such errors. In this paper we present a robust and computationally efficient corner ....

....Tracking and computation of structure and motion have been treated in the past as two separate problems. On one hand, traditional approaches to tracking have been based primarily on intensity correlation matching [2] with the more sophisticated approaches employing recursive Kalman filtering [11] for temporal consistency. Little attempt has been made to impose structural constraints on the features being tracked. On the other hand most approaches to structure and motion parameter estimation have assumed that reliable correspondences are already available [7, 15] The successes of these ....

[Article contains additional citation context not shown here]

I. D. Reid and D. W. Murray. Tracking Foveated Corner Clusters using Affine Structure. In Proc. Intl. Conf. Computer Vision, pages 76--83, 1993.

....Tracking and computation of structure and motion have been treated in the past as two separate problems. On one hand, traditional approaches to tracking have been based primarily on intensity correlation matching, with the more sophisticated approaches employing recursive Kalman filtering [6, 7] for temporal consistency. Little attempt has been made to impose structural constraints on the features being tracked [2, 8] On the other hand most approaches to structure and motion parameter estimation have assumed that reliable correspondences are already available [4, 9] The successes of ....

....being tracked [2, 8] On the other hand most approaches to structure and motion parameter estimation have assumed that reliable correspondences are already available [4, 9] The successes of these approaches are crucially dependent on the correctness of the assumed correspondences. Reid and Murray [6, 7] have developed a real time object tracker which uses a constant image velocity Kalman filter to establish the point correspondences across frames. They use the point correspondences obtained using their Kalman filter based correlation matcher to compute the 3D affine structure of the fixation ....

[Article contains additional citation context not shown here]

I. D. Reid and D. W. Murray. Tracking Foveated Corner Clusters using Affine Structure. In Proc. Intl. Conf. Computer Vision, pages 76--83, 1993.

....between left and right camera images to calculate depth information as well as using correlation between consecutive frames to perform tracking. The final form of tracking identifies specific feature locations in the image, and then performs correlation over those locations. Reid and Murray [RM93] demonstrate a robot head named Yorick that uses the feature correlation method to track objects. Reid and Murray s feature filter identifies locations in the image that look like corners. Corners are relatively easy to spot, and can be identified as corners over a wide range of poses and lighting ....

Ian D. Reid and David W. Murray. Tracking foveated corner clusters using affine structure. In Proceedings of the Fourth International Conference on Computer Vision, pages 76--83, Berlin, Germany, May 1993. IEEE Computer Society.

....for size constancy, we can approach the problem of zooming for scale change. Changing scale poses increased demands on target identification and localization because commonly used approaches such as cross correlation are not scale invariant. Alternatives include feature based tracking (e.g. Reid and Murray, 1993 ] or the use of adaptive correlation templates (e.g. Parry et al. 1995 ] We are currently investigating the latter approach. Target distance m sec 1.30 1.40 1.50 1.60 1.70 1.80 0.00 5.00 10.00 15.00 20.00 Focal length mm sec 19.00 20.00 21.00 22.00 23.00 24.00 0.00 5.00 10.00 15.00 ....

I.D. Reid and D.W. Murray. Tracking foveated corner clusters using affine structure. In Proc. of International Conference on Computer Vision, Berlin, Germany, May 1993, pp. 76--83.

....and discarded features to be removed. We thus employ the Tomasi Kanade algorithm on the first stereopair of our image sequence, and thereafter use the simplest version of the VSDF as presented in [16] Details may be found in [13] 5 Affine transfer The theory of affine transfer presented in [20, 21, 6] is a method of choosing a fixation point on a tracked object, so that image plane errors of the fixation point from the centre of the image(s) may be fed back to the motors controlling the camera(s) In this way a desired point on the object may be maintained at or near the centre of the image ....

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. 4th Int'l Conf. on Computer Vision, Berlin, pages 76--83, Los Alamitos, CA, 1993. IEEE Computer Society Press.

....for size constancy, we can approach the problem of zooming for scale change. Changing scale poses increased demands on target identification and localization because commonly used approaches such as cross correlation are not scale invariant. Alternatives include feature based tracking (e.g. Reid and Murray, 1993 ] or the use of adaptive correlation templates (e.g. Parry et al. 1995 ] We are currently investigating the latter approach. 2.4 NIST NIST is responsible for developing the vision processing platform, assisting in integrating University software onto the platform, and running the ....

I.D. Reid and D.W. Murray. Tracking foveated corner clusters using affine structure. In Proc. of International Conference on Computer Vision, Berlin, Germany, May 1993, pp. 76--83.

....simply by moving the attention window around in the full image. The more realistic looking method of moving the cameras to keep the attention window centered in the view will be discussed in Chapter 5. 3.1 Prior work Several techniques have been developed to track moving objects in real time. [ITI92, BCS92, Kam93, RM93] The systems tend to fall into three broad categories: the filter and follow method, the image correlation method, and the feature correlation method. Each involves one or another form of correlation, and each usually prefilters the image in some way to segment out the portion of the image that is ....

....in Chapter 4 3.1.3 Feature correlation The feature correlation technique uses a special filter to identify discrete point feature locations in the image that have some property. Tracking is performed by correlating the locations of the feature points from one frame to the next. Reid and Murray [RM93] demonstrate a robot head named Yorick that uses the feature correlation method to track objects. Reid and Murray s feature filter identifies locations in the image that look like corners. Corners are relatively easy to spot, and can be identified as corners over a wide range of poses and lighting ....

Ian D. Reid and David W. Murray. Tracking foveated corner clusters using affine structure. In Proceedings of the Fourth International Conference on Computer Vision, pages 76--83, Berlin, Germany, May 1993. IEEE Computer Society.

....of the model show that by detecting changes in a visual scene, the proposed model can track moving objects. There have been several pieces of work on tracking object motions from different aspects. For example, model based tracking ( 19] 25] 32] and [49] features tracking such as corners [39], contour tracking [6] lines tracking [2] etc. The work suggested here is a mechanism in which tracking can be done by a simple means. This model especially has the advantage that it is not restricted to work on motion only. 5.3 Directions for Future Research In this section, we will discuss ....

Reid, I. D. and Murray, D. W., "Tracking Foveated Corner Clusters Using Affine Structure", in proceedings of Fourth International Conference on Computer Vision, p. 76--83, 1993.

No context found.

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. 4th Int'l Conf. on Computer Vision, Berlin, pages 76--83, Los Alamitos, CA, 1993. IEEE Computer Society Press.

No context found.

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. ICCV'93, pages 76--83.

No context found.

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. ICCV'93, pages 76--83.

No context found.

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. 4th Int'l Conf. on Computer Vision, Berlin, pages 76--83, Los Alamitos, CA, 1993. IEEE Computer Society Press.

....in an approximately Euclidean frame. At any time the underlying affine structure can be made explicit, and we show the evolution of the convex hull as it is updated during tracking. Active and Real time Vision, Stereo, Shape recovery. Summary page Originality Affine transfer was introduced [24] as a method of exploiting the collective temporal coherence of a set of point features in the image which were individually likely to be short lived. That work demonstrated smooth transfer of fixation in a monocular camera. Robustness was provided by using descriptions of the fixation point in ....

....systems the ability to maintain fixation on the object. Tracking encompasses a broad spectrum of approaches. At one extreme is the 2D approach, where some image feature is tracked possibly a region (as in correlation [10, 11, 22] a contour (eg snakes [15, 30, 6, 8] or a point (eg corners [24]) Ideally the feature should be viewpoint invariant, but also robustly associated with the object of interest, a balance which is hard to achieve when there is no notion of what is being tracked. To provide robustness therefore, various degrees of model have been introduced, from 2D models (eg, ....

[Article contains additional citation context not shown here]

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. 4th Int'l Conf. on Computer Vision, Berlin, pages 76--83, Los Alamitos, CA, 1993. IEEE Computer Society Press.

....provides a linear batch algorithm which, if the same isotropic Gaussian noise is associated with each point, is optimal. McLauchlan [14] proposed an optimal recursive algorithm which could handle the birth and death of features. Transfer via affine structure was first used for planar tracking in [19], extended to monocular 3D transfer in [20] and yet further extended to stereo views in [4] the baseline yielding richer 3D information of the scene. However all this work assumed fixed camera intrinsics, and the last work was evidently limited by poor feature recovery and matching. Here we allow ....

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. ICCV'93, pages 76--83.

....better with respect to view point invariance and occlusion insensitivity, but at the cost of incorporating prior templates. The ideal method then would appear to be pointbased. An image corner feature is view point invariant, simple to extract and requires no prior model. However, as noted in [11, 12], although clusters of corners exhibit temporal coherence and longevity, individual corners are emphemeral and quite unsuitable for tracking over extended periods. When zoom is introduced, the situation becomes very much more difficult for all categories of method. Correlation now suffers ....

....same difficults as the contour method when corners fall off or enter the sides of image during zooming in and out, and appear or disappear because of scale effects. Recent work has demonstrated the active tracking of clusters of corner features using the method of affine transfer, both monocularly [11, 12] and stereoscopically [4] The method finesses the difficulty caused by the temporal instability of a single corner by, in the simplest case of 3D transfer, replacing the requirement to track one corner through all image frames by the less demanding requirement to track any four points across ....

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. 4th Int'l Conf. on Computer Vision, Berlin, pages 76--83, Los Alamitos, CA, 1993. IEEE Computer Society Press.

.... calibration of cameras typically achieved by viewing a well structured environment (e.g. a calibration grid) many researchers have begun to investigate the possibility of using algorithms which do not require calibration, are robust in the presence of calibration errors, or are self calibrating [1, 2, 3, 4, 5, 6]. Indeed this research has met with considerable success already, demonstrating that a tremendous amount may be achieved without the need for strict calibration. Stereo reconstruction is one area in which the apparent need for accurate calibration has plagued those attempting to build depth maps, ....

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. 4th Int'l Conf. on Computer Vision, Berlin, pages 76--83, Los Alamitos, CA, 1993. IEEE Computer Society Press.

....images is significantly more robust than that used in the Marvin system [22, 23] wherein odometry is encoded as an analogue signal at the start of the image data. To date, we have implemented six different behavioural reflexes on the system: coarse motion detection [6] corner tracking [24], looming (alarm detection) opto kinetic response and smooth pursuit [7] and a bright spot tracker [15] The last, although of limited visual interest, provides a convenient way to test the system, to debug the communications between the different sections and to demonstrate the performance of ....

....approximately time 1.8 seconds, the position demands begin to fluctuate considerably as the algorithm switches corners. The response remains stable only through use of the filtering described above. Details of this, and a more sophisticated corner tracker which eliminates this problem are given in [24]. 5 0 5 10 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 Seconds Degrees Left Vergence Axis 10 5 0 5 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 Seconds Degrees Elevation Axis Figure 12: Vergence and elevation response from the corner tracker using interpolated vision error signals. 6. CONCLUSIONS We have ....

I.D. Reid and D.W. Murray. Tracking foveated corner clusters using affine structure. Submitted to International Conference on Computer Vision, 1993.

....on parallel planes, ratios of volumes, and centroids. These are all useful sources of information for tasks which involve interaction with the environment: for instance, ratios can be used for the computation of time tocontact, and the centroid of a set of data points can be used for fixation [33] or grasping [19] Another affine invariant is the mid point locus between a set of points, a basic mechanism in path planning algorithms for navigation [22] Thus although it is traditional for path planning to be described in terms of Euclidean structure, many of the techniques will work ....

I. D. Reid and D. W. Murray. 1993. Tracking foveated corner clusters using affine structure. In Proc. 4th International Conference on Computer Vision, pages 76--83, Los Alamitos, CA, 1993. IEEE Computer Society Press.

....the underlying affine structure can be made explicit, and we show the evolution of the convex hull as it is updated during tracking. Keywords Active and Real time Vision, Stereo, Shape recovery. Submitted to 5th ICCV, December 5, 1994 3 Summary page Originality Affine transfer was introduced [24] as a method of exploiting the collective temporal coherence of a set of point features in the image which were individually likely to be short lived. That work demonstrated smooth transfer of fixation in a monocular camera. Robustness was provided by using descriptions of the fixation point in ....

....systems the ability to maintain fixation on the object. Tracking encompasses a broad spectrum of approaches. At one extreme is the 2D approach, where some image feature is tracked possibly a region (as in correlation [10, 11, 22] a contour (eg snakes [15, 30, 6, 8] or a point (eg corners [24]) Ideally the feature should be viewpoint invariant, but also robustly associated with the object of interest, a balance which is hard to achieve when there is no notion of what is being tracked. To provide robustness therefore, various degrees of model have been introduced, from 2D models (eg, ....

[Article contains additional citation context not shown here]

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. 4th Int'l Conf. on Computer Vision, Berlin, pages 76--83, Los Alamitos, CA, 1993. IEEE Computer Society Press.

....fewer features would have been required. However, this approximation was found to be insufficiently accurate for these sequences as the images exhibit non negligible perspective effects. Thus, although in the past the use of affine structure has proved fruitful for various active vision tasks [8, 14] such as tracking or visual servoing, it is not well suited to tackling quantitative measurements tasks unless the projection model truly is affine. The application we have presented is one of a wider class of problems (such as traffic monitoring) in which the ground plane trajectory of a target ....

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. 4th Int'l Conf. on Computer Vision, Berlin, pages 76--83, Los Alamitos, CA, 1993. IEEE Computer Society Press.

....on parallel planes, ratios of volumes, and centroids. These are all useful sources of information for tasks which involve interaction with the environment: for instance, ratios can be used for the computation of time to contact, and the centroid of a set of data points can be used for fixation [21] or grasping [14] Another affine invariant is the mid point locus between a set of points, a basic mechanism in path planning algorithms for navigation [16] Thus although it is traditional for path planning to be described in terms of Euclidean structure, many of the techniques will work ....

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. 4th International Conference on Computer Vision, pages 76--83, Los Alamitos, CA, 1993. IEEE Computer Society Press.

No context found.

I.D. Reid and D.W. Murray. Tracking foveated corner clusters using affine structure. In Proceedings of the 4th IEEE International Conference on Computer Vision, Berlin, Germany, 1993, pages 76--83, 1993.

No context found.

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. 4th Int'l Conf. on Computer Vision, Berlin, pages 76--83, Los Alamitos, CA, 1993. IEEE Computer Society Press.

No context found.

I. D. Reid and D. W. Murray. Tracking foveated corner clusters using affine structure. In Proc. 4th Int'l Conf. on Computer Vision, Berlin, pages 76--83, Los Alamitos, CA, 1993. IEEE Computer Society Press.

No context found.

I.D.Reid and D.W. Murray, "Tracking foveated corner clusters using affine structure". In Proceedings of the Fourth International Conference on Computer Vision, pages 76-83, Berlin, IEEE Computer Society, 1993.

No context found.

I. D. Reid and D. W. Murray, "Tracking Foveated Corner Clusters using Affine Structure", ICCV, 76 - 83, 1993.

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