N. Papanikolopoulos. "Selection of features and evaluation of visual measurements during robotic visual servoing tasks", Journal of Intelligent and Robotics Systems, 13:279--304, 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... a metric for the confidence in a motion estimate, so that estimates from regions of high confidence can be used to improve estimates in regions of low confidence [2] Feature tracking confidence measures have also been used in a visual servo control to increase the robustness of the servo control [3]. We characterize the accuracy of a feature tracking result by the accuracy of the location computed by the feature tracking. This characterization leads to the evaluation of the confidence of a feature tracking result for a more general purpose than that of accepting or rejecting the feature for ....

....images is defined as: SSD(u; v) X m;n2N [T (m; n) Gamma I(u m; v n) 2 ; 1) where T is the template image and I is the input image. The location (u; v) represents some location in the input image whose content is being compared to the content of the template. Papanikolopoulos [3] uses the SSD measure to generate tracking results that are then used for robotic visual servoing experiments. Anandan [2] and Singh and Allen [9] use this SSD metric for the computation of image flow. Often, this measure is not computed for the entire input image, but only for some search window ....

N. P. Papanikolopoulos, "Selection of features and evaluation of visual measurements during robotic visual servoing tasks," Journal of Intelligent and Robotic Systems, vol. 13, pp. 279--304, July 1995.

....its desired position and a possible trajectory in the image) 2. 1 Reaching or nearing a task singularity It is well known that the image Jacobian is singular if f is composed by the image of three points such that they are collinear, or belong to a cylinder containing the camera optical center [10]. Using more than three points generally allows one to avoid such singularities. However, whatever the number of points and their configuration, the image Jacobian may become singular during the visual servoing if image points are chosen as visual features. A simple example is described below. ....

N. Papanikolopoulos. "Selection of features and evaluation of visual measurements during robotic visual servoing tasks", Journal of Intelligent and Robotics Systems, 13:279--304, 1995.

....image surrounding the expected feature location. As this similarity measure is used on each pixel in a region, the scores make up what we term the SSD Surface (SSDS) The shape of the SSDS can tell us how well the template matches the image content surrounding the expected feature location [1] [29]. The amount of drop off in the SSDS, or equivalently the degree to which the template discriminates the most likely feature location from the surrounding area, is an important contributor to increasing the robustness of the tracking system as a whole. If the system can detect when an individual ....

N. P. Papanikolopoulos. Selection of features and evaluation of visual measurements during robotic visual servoing tasks. Journal of Intelligent and Robotic Systems, pages 279-- 304, July 1995.

....in their measurement, which enables applications to weight features appropriately, depending on the associated con dence level and on the application s needs. In this paper, we consider only the case of feature trackers that use the sum of squareddi erences (SSD) correlation method [21] [19] [1] In SSD based feature tracking, a feature template is compared to portions of an image to locate that feature. This comparison uses a similarity metric to rate the similarity of the template and the image patch. The image region found to be the most similar to the template is typically taken ....

....from regions of high con dence are used to improve estimates in regions of low con dence. To facilitate this, a metric for the con dence in a motion estimate is developed. Finally, feature tracking con dence measures have been used in visual servo control to increase the robustness of the control [19]. In each of these cases, the uncertainty characterizations are used only to reject or accept features. In contrast, the approach that we present here gives a quantitative evaluation of uncertainty that can be used to weight feature measurements in proportion to their reliability. In our ....

[Article contains additional citation context not shown here]

N. P. Papanikolopoulos. Selection of features and evaluation of visual measurements during robotic visual servoing tasks. Journal of Intelligent and Robotic Systems, 13(3):279-304, July 1995.

....work uses a robotic arm exercising three internal degrees of freedom. This work demonstrates that object tracking with explicit models of objects with complex geometry is feasible, even for objects with many internal degrees of freedom. Previous work has assumed a constant set of visible features [10], or the use of features with simple primitives (edges or corners) 5] Since we use realistic object and imaging models to compute the expected visibility of features, we can accommodate a set of features distributed about the object such that several will be visible in most configurations of the ....

.... fit of that feature to the surrounding image is returned, in the form of an SSD surface [1] The local shape of the SSD surface can tell in which directions the template is a good match with the local image content, and in which directions the template differs greatly from the local image content [10]. If we interpret the feature location as a two dimensional random vector, the surface returned by an SSD measurement can be normalized and treated as a probability measure on the image plane position of the feature. Since the matching properties of the SSD measurement are only valid within some ....

[Article contains additional citation context not shown here]

N. P. Papanikolopoulos. Selection of features and evaluation of visual measurements during robotic visual servoing tasks. Journal of intelligent and robotic systems, pages 279--304, July 1995.

....that the shape and appearance of the object being tracked are known. This assumption, also not an uncommon one [11] allows us to generate templates for the predicted appearance of features of interest in a given configuration. An alternative to this assumption is to use features gathered on line [12], which may help to ensure the quality of the features tracked, since feature templates exactly match previous feature appearance. However, this method destroys any a priori information about the geometric relationship of the individual features to the object. 2 Observe Image Joint Angles Figure ....

....structure and choosing feature points with strong invariant properties. An invariant property is any property that does not vary with respect to changes in nuisance parameters. A nuisance parameter is any parameter not of interest in the current situation. Shi and Tomasi [29] and Papanikolopoulos [12] independently proposed confidence measures to use for the purpose of feature selection. 2.3 Region Tracking A common special case of feature tracking as described above occurs when the features to be tracked are regions in an image. A region is often defined as a maximal homogeneous image patch, ....

[Article contains additional citation context not shown here]

N. P. Papanikolopoulos, "Selection of features and evaluation of visual measurements during robotic visual servoing tasks," Journal of Intelligent and Robotic Systems, vol. 13, pp. 279--304, July 1995. 180

....tracked is known, and that an appearance model of the object is available. This assumption, not an uncommon one, allows us to generate templates for the predicted appearance of features of interest in a given configuration. An alternative to this assumption is to use features gathered on line [10], which may help to ensure the quality of the features tracked, since feature templates exactly match previous feature appearance. However, this method fails to exploit any a priori information about the geometric relationship of the individual features to the object. We will relax this assumption ....

N. P. Papanikolopoulos. Selection of features and evaluation of visual measurements during robotic visual servoing tasks. Journal of Intelligent and Robotic Systems, 13(3):279--304, July 1995.

....be reached owing to the existence of unrealizable image motions. 2. 1 Reaching or nearing a task singularity It is well known that the image Jacobian is singular if s is composed by the image of three points such that they are collinear, or belong to a cylinder containing the camera optical center [3,19,22]. Using more than three points generally allows us to avoid such singularities. However, we now demonstrate by a concrete example that, whatever the number of points and their configuration, the image Jacobian may become singular during the visual servoing, if image points are chosen as visual ....

N. Papanikolopoulos. Selection of features and evaluation of visual measurements during robotic visual servoing tasks. Journal of Intelligent and Robotics Systems, 13:279--304, 1995.

.... for automatic selection of features and on line evaluation of visual measurements (useful in cases of sudden occlusion of features) is presented in [14] In order to solve for the manipulator control input, it can be shown that at least three feature points which are not collinear are needed [17]. In other words, less than three feature points do not provide enough measurements in order to reliably compute the manipulator control input. The state space model for M (M 3) feature points can be written as where A=H=I 2M , E=TI 2M , x(k) R 2M , d(k) R 2M , u(k) R 6 , v(k) R 2M . ....

N.P. Papanikolopoulos and P.K. Khosla, "Selection of Features and Evaluation of Visual Measurements for 3-D Robotic Visual Tracking" in Proc. of the 1993 IEEE International Symposium on Intelligent Control, pp. 320-325, August 25-27, 1993.

No context found.

N. Papanikolopoulos. "Selection of features and evaluation of visual measurements during robotic visual servoing tasks", Journal of Intelligent and Robotics Systems, 13:279--304, 1995.

No context found.

N. P. Papanikolopoulos and P. K Khosla. Selection of features and evaluation of visual measurements for 3-d robotic visual tracking. Int. Symp. on Intelligent Control., pages 320--325, August 1993.

No context found.

N. P. Papanikolopoulos, "selection of features and evaluation of visual measurements during robotic visual servoing tasks," Journal of Intelligent and Robotic System, 13(3), 279-304, 1995.

No context found.

N. Papanikolopoulos. "Selection of features and evaluation of visual measurements during robotic visual servoing tasks", Journal of Intelligent and Robotics Systems, 13:279--304, 1995.

No context found.

N. P. Papanikolopoulos, Selection of features and evaluation of visual measurements during robotic visual servoing tasks, Journal of Intelligent and Robotic Systems 13 (3) (1995) 279--304.

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