M. Black and P. Anandan. Constraints for the early detection of discontinuity from motion. AAAI, pp. 1060-- 1066, 1990.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....difficult to recover. As a result, the first class of algorithms are operating in regions of the image where the optical flow is least accurate. 2. 1 Motion Field Segmentation Early work on segmentation via motion looked for discontinuities in one or both components of the displacement field [SU87, TMB85, BA90, Bla92]. Since under general perspective projections the motion field is continuous as long as the depth of the viewed surface is continuous, discontinuities in the flow field signal depth discontinuities. Unfortunately, the flow field at discontinuities is difficult to recover. For example, optical flow ....

M. Black and P. Anandan. Constraints for the early detection of discontinuity from motion. In Proceedings of the National Conference on A.I., pages 1060--1066, Boston, 1990.

....determine which portions of the image are moving coherently, and thus should be integrated together. A common approach to motion grouping has been to use an edge field in combination with a velocity field, and use a smoothing mechanism that does not integrate information across edge boundaries [4]. This type of line process approach works well when the scene contains only simple occlusion, in which each object projects to a relatively large, connected region in the image. But in more complicated situations, such as when the projected regions of an object are very small and disjoint (i.e. ....

M. Black and P. Anandan. Constraints for the early detection of discontinuity from motion. In Proc Eighth National Conf on Artificial Intelligence, pages 1060--1066, July 1990.

....30] assuming a stimulus is a linear superposition of independent motion signals; kinetic occlusion cannot be considered as such. ffl Robust estimation. Motion estimation methods have been modified to be less sensitive to violations of underlying assumptions of smoothness and data conservation [3, 4, 31]. By discounting deviations from underlying assumptions, occlusions are viewed as an anomalous process that should be ignored. ffl Multiple motions globally. Parametrized estimates of local motion are typically clustered together in velocity space; these clusters can be used to describe global ....

M. J. Black and P. Anandan. Constraints for the early detection of discontinuity from motion. In Proc. National Conf. on Artificial Intelligence, AAAI-90, pages 1060--1066, Boston, MA, 1990.

....values (although he uses the absolute value of the intensity difference rather than the square) A number of authors have addressed the motion segmentation problem. Spoerri and 6 More sophisticated tests could be done, but this has worked fairly well to date. Ullman [35] and Black and Anandan [9], for example, use the shape of the local correlation surface to find moving objects. Fran cois and Bouthemy s work [15] is quite related to our tracking algorithm, but their emphasis is not on real time vision. Model based tracking appears to have originated with Gennery [16] Both Lowe and ....

Michael Black and Padmanabhan Anandan. Constraints for the early detection of discontinuity from motion. In Proceedings of AAAI-90, Boston, MA, pages 1060--1066, 1990.

No context found.

M. Black and P. Anandan. Constraints for the early detection of discontinuity from motion. AAAI, pp. 1060-- 1066, 1990.

No context found.

M. Black and P. Anandan. Constraints for the early detection of discontinuity from motion. AAAI, pp. 1060-- 1066, 1990.

No context found.

M. J. Black and P. Anandan. Constraints for the early detection of discontinuity from motion. Proc. AAAI, pp. 1060--1066, 1990.

No context found.

M. J. Black and P. Anandan. Constraints for the early detection of discontinuity from motion. Proc. AAAI, pp. 1060--1066, 1990.

....but it has often been noted that these methods are sensitive to the accuracy of the optical flow and that accurate optical flow is hard to estimate without prior knowledge of the occlusion boundaries. Other methods have focused on detecting occlusion from the structure of a correlation surface [3], or of the spatiotemporal brightness pattern [7, 9, 17] Still others have used the presense of unmatched features to detect dynamic occlusions [16] None of these methods explicitly model the image motion present at a motion feature, and have not proved sufficiently reliable in practice. For ....

M. Black and P. Anandan. Constraints for the early detection of discontinuity from motion. AAAI, pp. 1060-- 1066, 1990.

....more complex features such as moving bars. Another approach uses filters tuned to the local spatiotemporal image structure of a motion discontinuity [6, 8, 11] These methods model image structure over time rather than the motion field. For low level motion based detection, Black and Anandan [2] look for multiple peaks in a sum of squared difference surface as evidence of an occlusion disocclusion boundary. Although they detect occlusions, these methods do not explicitly model the image motion present at a motion feature. In [4] the authors propose learning explicit linear models of ....

M. J. Black and P. Anandan. Constraints for the early detection of discontinuity from motion. Proc. AAAI, pp. 1060--1066, 1990.

.... to modeling step edges in static scenes by learning a parameterized model from examples of edges [14] It differs from previous attempts to detect motion discontinuities that applied edge detectors to optical flow, checked for bimodality in local flow estimates, or used energy based methods [4, 15, 17]. 3 Learning Parameterized Flow Models Learning a parameterized model for a particular class of motions requires that we have a training set of flow fields containing representative samples of the class. For relatively simple classes such as motion discontinuities we can generate this training ....

M. J. Black and P. Anandan. Constraints for the early detection of discontinuity from motion. AAAI'90, pp. 1060--1066.

No context found.

M. Black and P. Anandan. Constraints for the early detection of discontinuity from motion. In Proc Eighth National Conf on Artificial Intelligence, pages 1060--1066, July 1990.

No context found.

M. Black and P. Anandan. Constraints for the early detection of discontinuity from motion. In Proc Eighth National Conf on Artificial Intelligence, pages 1060--1066, July 1990.

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