A. J. Davison and D. W. Murray. Mobile robot localisation using active vision. In Proceedings of the 5th European Conference on Computer Vision, Freiburg, pages 809--825, 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....images and the features found. They introduce inaccuracy in both the landmarks position as well as the least squares estimation of the robot position. In stochastic mapping, a single lter is used to maintain estimates of landmark positions, the robot position and the covariances between them [4], with high computational complexity. In more recent work, we have employed a Kalman Filter [1] for each database SIFT landmark which now has a 3x3 covariance matrix for its position, assuming the independence of landmarks. When a match is found in the current frame, the covariance matrix in the ....

A.J. Davison and D.W. Murray. Mobile robot localisation using active vision. In Proceedings of Fifth European Conference on Computer Vision (ECCV'98) Volume II, pages 809-825, Freiburg, Germany, June 1998.

....to the problem of active visual perception in real time problem solving tasks such as air traffic control. Active vision refers to systems that not only sense, but also interact with the world during sensing, by focusing attention, processing selectively, choosing where to look, etc. 9] [10], 33] We want to look at reinforcement learning algorithms for learning the COMDPs. During the early phases of learning, the controller can observe the entire state of the system and acquire an accurate model. 6.3 Schedule ffl 1999: work on approximation algorithms for COMDPs do a ....

A.J. Davison and D.W. Murray. Mobile robot localisation using active vision. Proceedings of the European Conference on Computer Vision, 1998. 18

....Tracking has been studied extensively in the computer vision literature, both because of its intrinsic interest and because of the large number of applications. For example, autonomous robots may need to be able to follow objects in their environment (Reid and Murray, 1996; Pahlavan et al. 1993; Davison and Murray, 1998; Murray et al. 1993; Espiau et al. 1992) one commonly studied special case of this concerns autonomous guided vehicles for driving on roads, which must track the features of the road (Dickmanns, 1992; Crisman, 1992) and also other moving vehicles (Smith, 1995) Static systems may also be used ....

Davison, A. J. and Murray, D. W. (1998). Mobile robot localisation using active vision. In Submitted to the 5th European Conference on Computer Vision, Freiburg.

.... so the alternative Kalman filter may be preferable where a good motion model is available, and where motion reconstruction rather than scene reconstruction is the major aim, so that a smaller representative subset of features may be processed, circumventing the computational complexity problems [9], or else scene reconstruction may be eliminated altogether [46] We have chosen a single current application of the VSDF in order to provide a context for the description of the general VSDF algorithm. This is the reconstruction of 3D features and surfaces using structure from motion methods. ....

....Kalman filter update for our state vector 9, even if it could be computed, would take hours. We need an alternative approach, that approximates as closely as possible the accuracy of the Kalman approach, while reducing the update time. Note however that Broida Chelappa [7] and Davison Murray [9] successfully applied the Extended Kalman filter (EKF) to problems involving a small number of structure vectors. 2. The structure from motion problem is usually formulated so that all the camera motion estimates are computed without any dynamical model, i.e. the camera position at any given time ....

A.J. Davison and D.W. Murray. Mobile robot localisation using active vision. In Proc. 5th European Conf. on Computer Vision, Freiburg. Springer-Verlag, June 1998.

....circumvents problems with the previous method that became apparent when incomplete matching data was used. In other related work, Broida et al. 2] applied the Kalman filter in a straightforward way to the motion and structure estimation problem, using a finite dynamical model. Davison Murray [5] also applied the Kalman 1 Here recursive means that a large linear system is partitioned so that a smaller system can be solved, which is itself recursively partitioned, etc, NOT in the sense of a recursive filter. Thus we use recursive partitioning in both the batch and recursive stages. ....

A. Davison and D. Murray. Mobile robot localisation using active vision. In Proc. 5th European Conf. on Computer Vision, Freiburg. Springer-Verlag, June 1998.

No context found.

A. J. Davison and D. W. Murray. Mobile robot localisation using active vision. In Proceedings of the 5th European Conference on Computer Vision, Freiburg, pages 809--825, 1998.

No context found.

A. J. Davison and D. W. Murray, "Mobile robot localisation using active vision," in Proceedings of the 5th European Conference on Computer Vision, Freiburg, 1998, pp. 809--825.

No context found.

A. J. Davison and D. W. Murray. Mobile robot localisation using active vision. In Proceedings of the 5th European Conference on Computer Vision, Freiburg, pages 809--825, 1998.

No context found.

A. J. Davison and D. W. Murray. Mobile robot localisation using active vision. In Proceedings of the 5th European Conference on Computer Vision, Freiburg, pages 809--825, 1998.

No context found.

A.J. Davison, D.W. Murray, Mobile robot localisation using active vision, in: Proceedings of the Fifth European Conference on Computer Vision, Freiburg, Germany, 1998, pp. 809--825.

No context found.

A. J. Davison and D. W. Murray, "Mobile robot localisation using active vision," in Proceedings of the 5th European Conference on Computer Vision, Freiburg, 1998, pp. 809--825.

No context found.

A. J. Davison and D. W. Murray. Mobile robot localisation using active vision. In Proceedings of the 5th European Conference on Computer Vision, Freiburg, pages 809--825, 1998.

No context found.

A. J. Davison and D. W. Murray. Mobile robot localisation using active vision. In Proceedings of the 5th European Conference on Computer Vision, Freiburg, pages 809-825, 1998.

....from ground truth in a systematic way, as can be seen in the experiments of the authors referenced above. They are not able to produce sensible estimates for long runs where previously seen features may be revisited after periods of neglect, an action that allows drifting estimates to be corrected [8]. 2.3 Propagating Coupled Estimates To work with coupled estimates, it is necessary to propagate not only each estimated quantity and its uncertainty, but also how this relates to the uncertainties of other estimates. Generally, a group of uncertain quantities is represented by a probability ....

....cases the Extended Kalman Filter provides an approximation which in general has been found to perform very well. Called Stochastic Mapping in its rst correctly formulated application to robot map building [19] the EKF has been implemented successfully in di erent scenarios by other researchers [3, 5, 6, 8, 10, 15]. Its main weakness compared to the Monte Carlo methods is its inability to represent multi modal distributions where an estimate has two or more peak values that are most likely, with unlikely regions in between. The Monte Carlo methods gain greatly in robustness through this ability. In rst ....

[Article contains additional citation context not shown here]

A. J. Davison and D. W. Murray. Mobile robot localisation using active vision. In Proceedings of the 5th European Conference on Computer Vision, Freiburg, pages 809-825, 1998.

.... freedom and foveated lenses which provide high resolution in the image centre and low resolution in the periphery [5] 2 Localisation Using Active Vision In this section we will describe the localisation process of a single robot equipped with a vision system, closely following the approach of [2], before moving on to the multiple robot case in Section 3. 2.1 Active Vision In active approaches to sensing, sensor or information processing resources are directed purposively to regions of current interest in a scene, rather than being used to acquire and process data uniformly. In vision, ....

.... tasks such as steering around a known obstacle: the obstacle is tracked and a simple law can be used to control the robot s steering with respect to the viewing angle [7] However, there have only been a few attempts to apply it to more long term navigation tasks such as map building and using [2]. This is surprising since it is the main tool used in human navigation: as we move around our environment, our eyes constantly change their fixation point to look for of landmarks, check for obstacles or pick out headings. Attention must be divided between these important tasks as required, and ....

[Article contains additional citation context not shown here]

A. J. Davison and D. W. Murray. Mobile robot localisation using active vision. In Proceedings of the 5th European Conference on Computer Vision, Freiburg, pages 809--825, 1998.

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