Q. Zheng and R. Chellappa, "Automatic feature point extraction and tracking in image sequences for arbitrary camera motion," Int. J. Comput. Vis., vol. 15, pp. 31--76, 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....similar over time it is mostly applied to greylevel or color image sequences. In [19] and [113] template matching is employed for vehicle tracking in greylevel sequences. Region based matching of height maps is applied in [5] Many other approaches are based on finding feature correspondences. In [117], for example, Zheng et al. describes an intensity based approach to feature tracking. Problems caused by feature location quantization and perspective deformations are handled by using subpixel accuracy matching and weighted correlation. Also in [83] feature correspondences are determined to ....

Q. Zheng and R. Chellappa. Automatic feature point extraction and tracking in image sequences for arbitrary camera motion. International Journal of Computer Vision, 15:31-- 76, 1995.

....visual tracking approaches assume much narrower and incomplete models of the gesticulating person. Feature based approaches assume that the users gesture movements give rise to image features that can be detected and used for tracking. Common visual features used are contours [61 63, 45] points [64 67], color [68, 69] and motion [70] Finally image content itself can directly serve as image features [71 74] Contour based approaches suffer from the requirement that they usually require some form of more detailed model of the target to be tracked. This makes the approaches often unsuitable ....

Q. Zheng and R. Chellappa, "Automatic feature point extraction and tracking in image sequences for arbitrary camera motion," International journal of computer vision, vol. 15, pp. 31-76, 1995.

....temporal model of activity, individual object blobs generated by motion detection are tracked over time by matching them between frames of the video sequence. Among the many approaches to tracking are model based matching [9] 42] 43] image contour matching [44] and image region matching [45] [46]. Multiple potential matches typically arise, which can be disambiguated using statistical data association techniques [47] 48] or by imposing smooth trajectory motion models using Kalman filters [49] Our approach lies squarely in the image region matching camp. Given a moving object region in ....

Q. Zheng and R. Chellappa, "Automatic feature point extraction and tracking in image sequences for arbitrary camera motion," Int. J. Comput. Vis., vol. 15, pp. 31--76, June 1995.

....and particular features (points of Moravec [5] edge points, corners, junctions, etc) There are numerous algorithms to match such interest points between views. Among these well known algorithm we can mention the correlation technique, methods of optimization, and the dynamic programming [8,11]. In this paper, our focus here is on a couple of rectified images, obtained from a fixed polyhedral scene. Paragraph 2 describes briefly the correlation based technique and the dynamic programming method. In paragraph 3, the main improvement and process associated with the utilization of the ....

Zheng Q., Chellappa R.., Automatic feature point extraction and tracking in image sequences for arbitrary camera motion. Int. Journal of Computer Vision, 15(1/2):31-76,1995. 4 a) b) c) d) e) f)

....matching free point features through the sequence. Since matching occurs during boundary definition, information about the object such as the foreground color and object shape in the vicinity of the point being matched is not explicitly available. Though point matching techniques abound [172,147,185], many of them fail to choose the correct match when multiple high confidence possibilities are available or when local properties change. Without some higherlevel knowledge of the object and how it moves, it is difficult to provide reliable matches of edge points in a dynamic system. ....

Q. Zheng and R. Chellappa, "Automatic feature point Extraction and Tracking in Image Sequences for Unknown Camera motion," IEEE Proc. of the Fourth International Conference on Computer Vision (ICCV `93), pp. 335-339, Berlin, Germany, May 1993.

....Concerning corner tracking, Shapiro et al. 1] detect and track corners individually in an algorithm originally aimed at applications such as videoconferencing. Smith and Brady [2] track a large set of corners and use the results in a ow based segmentation algorithm. Zheng and Chellappa [3] have studied feature tracking when compensating for camera motion, and Gee and Cipolla [4] track locally darkest points with applications to pose estimation. In contour tracking, snakes can be used to track moving, deforming image features [5, 6] and such an approach is applied to estimate ....

Q. Zheng and R. Chellappa, \Automatic feature point extraction and tracking in image sequences for arbitrary camera motion," International Journal of Computer Vision, vol. 15, no. 1, pp. 31-76, 1995.

....N corresponds to the number of features to be tracked. Each zone is searched from top to bottom, and the topmost feature is selected for tracking. The features selected in frame f t Gamma1 can be tracked to frame f t by a multi resolution refinement scheme using parameter estimation proposed in [21]. Although this process is able to produce very good estimates of the motion parameters, it is computationally expensive since it uses a weighted correlation scheme to determine the best feature matches. We use another similarity measure which is given by the SSD over local windows (SSD windows) ....

....can handle can be very large even for small values of s. 2.1. 5 Subpixel Matching After the grid to grid matches are obtained from the hierarchical search, displacements with subpixel accuracy can be easily computed for the finest resolution level of the pyramid using a differential method [18, 21]. Subpixel accuracy is necessary to eliminate the quantization error introduced when the images are digitized. If a feature P t (u; v) has offset (ffix; ffiy) relative to P t Gamma1 (u; v) assume they were tracked and registered so that the translation (ffix; ffiy) is very small) i.e. ....

Q. Zheng and R. Chellappa. Automatic feature point extraction and tracking in image sequences for unknown camera motion. International Journal of Computer Vision, 15:31--76, 1995. 44

....frames. An excellent recent survey of optical flow computation schemes was performed by Barron et al. 9] Motion recovery is an active research area in computer vision and many approaches have been taken. One approach is to rely on the matching of feature points between consecutive frames [10]. However, these approaches tend to be computationally expensive and unreliable in the presence motion blur. Another approach is based on the computation of the optical flow [11] for every pixel. These methods are very computationally expensive and difficult to implement in real time. They are ....

Q. Zheng and R. Chellappa, `Automatic Feature Point Extraction and Tracking in Image Sequences for Unknown Camera Motion', Proc. of ICCV93, pp. 335-340.

.... features (such as corners) have the advantage that the full optical flow is known at every measurement position, because they do not suffer from the aperture problem effect (a discussion on this subject can be found elsewhere [5] Work on tracking two dimensional features can be found elsewhere [6 10]. Robust tracking means automatically detecting unreliable matches, or outliers, over an image sequence (see elsewhere [11] for a survey of robust methods in Computer Vision) Recent examples of such robust algorithms include reference 12, which identifies tracking outliers while estimating the ....

Zheng Q, Chellappa R. Automatic feature point extraction and tracking in image sequences for arbitrary camera motion. International Journal of Computer Vision 1995; 15(15):31--76

....points in the scene. Much because of this, corner tracking has received a lot of attention during the last ten years and we will here mention some works. Shapiro et al. 1992b) detected and tracked corners individually in an algorithm originally aimed at applications such as video conferencing. (Zheng and Chellappa, 1995) studied corner tracking when compensating for camera motion. Both these works based the matching on cross correlation of corner patches. Smith and Brady, 1995) tracked a large set of corners and used the results in a flow based segmentation algorithm. Here, the distance between the candidates ....

Zheng, Q. and Chellappa, R. (1995). Automatic feature point extraction and tracking in image sequences for arbitrary camera motion, Int. J. of Computer Vision 15(1): 31--76.

....chose 38 feature points in this image and obtained their displacement values with respect to the right image using Anandan s algorithm[1] these are shown as vectors superimposed on the image in Fig. 8. The feature points may be extracted using automated feature detection algorithms such as [26]. Using all 38 points in the linear system, the singular values of the matrices D and [Db] of KGP are listed in Table 1. From the table, oe min (D) 0:01689 and oe min (Db) 0:01667, respectively. Thus, the ratio value is 1.01, which is very close to 1. On the other hand, using only the 21 ....

Q. Zheng and R. Chellappa. "Automatic feature point extraction and tracking in image sequences for unknown camera motion". In Proc. ICCV. IEEE, 1993.

....matching a set of image features (e.g. points, edges or regions) present in one frame with features found in the next frame. Image points are typically matched using conventional correlation, where an image patch extracted from the previous frame is compared with a search area in the current frame [1, 2, 3]. Point correlation has the advantage of being robust and can be implemented for real time applications. However, the problem associated with correlation is that, over a long image sequence, small tracking errors may accumulate to cause tracked points to migrate from their initially designated ....

....whose shape, size or orientation changes across consecutive frames produces an error in the correlation. Although the error is small given a short time step between two consecutive frames, the accumulation of errors over a large number of frames cause tracked points to migrate from the target area [2, 3]. The tracking of multiple points allows the motion parameters of a moving camera to be estimated. For the purpose of guiding a missile or aircraft towards a stationary object only a subset of the motion parameters are required for navigation. The four dominant incremental guidance parameters are ....

Q. Zheng and R. Chellappa, "Automatic Feature Point Extraction and Tracking in Image Sequences for Arbitrary Camera Motion", International Journal of Computer Vision, 15, pp. 31-76, 1995.

....matrix. s , T , and R the corrections for s, T, and R, respectively. 4 I. Introduction Image registration is an important technique for a great variety of applications such as aerial image analysis [1] 2] 3] stereo vision [4] 5] automated cartography [6] motion analysis[7], 8] and the recovery of the 3 D characteristics of a scene [9] There are two tasks which need to be handled during an image registration process. They are feature selection and correspondence establishment. Typically, feature points can be selected by manual or automatic methods [1] 6] ....

Q. Zheng and R. Chellappa, "Automatic feature point extraction and tracking in image sequences for arbitrary camera motion," I. J. Computer Vision, vol. 15, pp. 31--76, 1995.

.... of two dimensional features (such as corners) have the advantage that the full optical flow is known at every measurement position, because they do not suffer from the aperture problem effect (a discussion on this subject can be found in [24] Works on tracking of two dimensional features include [13, 1, 6, 18, 26]. Robust tracking means detecting automatically unreliable matches, or outliers, over an image sequence (see [14] for a survey of robust methods in computer vision) Recent examples of such robust algorithms include [23] which identifies tracking outliers while estimating the fundamental matrix, ....

Q. Zheng and R. Chellappa. Automatic feature point extraction and tracking in image sequences for arbitrary camera motion. International Journal of Computer Vision, 15(15):31--76, 1995. Improving Feature Tracking with Robust Statistics

....edges or regions) present in one frame with features found in the next frame. Image points are typically matched using conventional correlation, where an image patch extracted from the previous frame is compared with a search area in the current frame to find a point of greatest similarity [1, 2, 3]. Point correlation has the advantage of being straight forward to implemented for a wide range of real time applications. However, the problem associated with correlation is that, over a long image sequence, small tracking errors may accumulate to cause tracked points to migrate from their ....

....However, the problem associated with correlation is that, over a long image sequence, small tracking errors may accumulate to cause tracked points to migrate from their initially designated areas. image patch extracted from the previous frame is compared with a search area in the current frame [1, 2, 3]. The advantage of point correlation is that it is robust and can be implemented for real time applications. However, the problem associated with correlation is, that over a large image sequence, small tracking errors may accumulate to cause tracked points to migrate from their initially ....

[Article contains additional citation context not shown here]

Q. Zheng and R. Chellappa, "Automatic Feature Point Extraction and Tracking in Image Sequences for Arbitrary Camera Motion", International Journal of Computer Vision, 15, pp. 31-76, 1995.

....of larger window sizes, for reliability purposes, but the weighting function favors the central portion of the window, at the expense of the outer regions. The influence of perspective effects are thus minimized. Similar approaches have been employed by other researchers (see, Mori et al. 1973; Zheng and Chellapa, 1993; Xin, 1995; Tao, 1997, for example) A window size of 11 x 15 is used. Epipolar geometry Since the sensors are mounted in a fixed configuration on the platform, accurate relative orientation between imaging sensors is known. This information is independent of the absolute sensor position, and is ....

Q. Zheng and R. Chellapa. Automatic feature point extraction and tracking in image sequences for unknown camera motion. In Proceedings of the Fourth International Conference on Computer Vision, pages 335--339, Berlin, Germany, 1993.

....of velocities of 3D surface points onto the imaging plane of a perspective camera. The importance of motion in visual processing cannot be understated: in particular, approximations to image motion may be used to estimate 3D scene properties and motion parameters from a moving visual sensor [19, 28, 29, 40, 49, 48, 1, 5, 36, 20, 50, 52, 32, 18, 15, 21], to perform motion segmentation [7, 38, 43, 34, 45, 13, 23, 8, 2, 44, 14] to compute the focus of expansion and time to collision [42, 39, 46, 22, 47, 9] to perform motioncompensated image encoding [10, 12, 33, 35, 37, 51] to compute stereo disparity [3, 11, 24, 26] to measure blood flow and ....

Q. Zheng and R. Chellappa. Automatic feature point extraction and tracking in image sequences for unknown image motion. In Proceedings of ICCV, pages 335--339, Berlin, Germany, May 1993.

.... both increased computation efficiency and the ability to find better solutions (escape from local minima) Tracking individual features (corners, points, lines) in images has always been alternative to iconic (pixel based) optic flow techniques [Dreschler and Nagel, 1982; Sethi and Jain, 1987; Zheng and Chellappa, 1992] This has the advantage of requiring less computation and of being less sensitive to lighting variation. The algorithm presented in this paper is closely related to patchbased feature trackers [Lucas and Kanade, 1981; Rehg and Witkin, 1991; Tomasi and Kanade, 1992] In fact, our general motion ....

Q. Zheng and R. Chellappa. Automatic Feature Point Extraction and Tracking in Image Sequences for Arbitrary Camera Motion. Technical Report CAR-TR-628, Computer Vision Laboratory, Center for Automation Research, University of Maryland, June 1992.

....I x e k;l P k;l I y e k;l 3 5 : 3) The matrix on the left hand side is often referred to as the Hessian of the system, and encodes the relative certainties in the flow estimates. 1 For tracking long sequences, a combination of searchbased correlation and the differential method can be used [Zheng and Chellappa, 1992]. 1 More formally, the covariance matrix of the flow estimate is proportional to the inverse Hessian [Szeliski, 1989] 3 Spline based image registration 3 The basic correlation technique works well when the motion is mostly (locally) translational between frames, and when there are no large ....

Q. Zheng and R. Chellappa. Automatic Feature Point Extraction and Tracking in Image Sequences for Arbitrary Camera Motion. Technical Report CAR-TR628, Computer Vision Laboratory, Center for Automation Research, University of Maryland, June 1992.

....As examples of feature tracking, Shapiro [1] detects and tracks corners individually in an algorithm originally aimed at applications such as videoconferencing. Smith and Brady [2] track a large set of corners and use the results in a flow based segmentation algorithm. Zheng and Chellappa [3] track features while compensation for camera motion. Cipolla and Blake [4] estimate time to contact by using snakes to track deforming image features, and Koller et al. 5] track combined motion and grey level boundaries in traffic surveillance. For edge based tracking approaches, see Faugeras ....

Q. Zheng and R. Chellappa, "Automatic feature point extraction and tracking in image sequences for arbitrary camera motion," IJCV, vol. 15, no. 1, pp. 31--76, 1995.

.... (such as corners) have the advantage that the full optical flow is known at every measurement position, because they do not suffer from the aperture problem effect (a discussion on this subject can be found in [149] 86 Feature Tracking Works on tracking of two dimensional features include [89, 8, 23, 127, 170]. Robust tracking means detecting automatically unreliable matches, or outliers, over an image sequence (see [103] for a survey of robust methods in computer vision) Recent examples of such robust algorithms include [144] which identifies tracking outliers while estimating the fundamental ....

Zheng, Q., and Chellappa, R. Automatic feature point extraction and tracking in image sequences for arbitrary camera motion. International Journal of Computer Vision 15, 15 (1995), 31--76.

....geometry of the scene or the motion of the visual sensor are partially or completely pre determined. The importance of motion in visual processing cannot be understated: approximations to image motion may be used to estimate 3 d scene properties and motion parameters from a moving visual sensor [63, 90, 92, 120, 149, 148, 3, 15, 112, 65, 162, 164, 98, 51, 45, 72], to perform motion segmentation [23, 115, 124, 103, 137, 41, 75, 28, 9, 125, 44] to compute the focus of expansion and time to collision [123, 117, 140, 74, 142, 29] to perform motion compensated image encoding [34, 40, 100, 104, 114, 163] to compute stereo disparity [12, 38, 76, 84] to ....

Q. Zheng and R. Chellappa. Automatic feature point extraction and tracking in image sequences for unknown image motion. In Proceedings of ICCV, pages 335--339, Berlin, Germany, May 1993.

....The filter method attempts to increase the signal to noise ratio in the difference image and thus increase the accuracy of the motion detection. 86] Thus affine motion models are widely used in computer vision for motion segmentation and tracking (i.e. Cipolla Blake 92; Murray et al. 93; Zheng Chellappa 93] We assume that the apparent image motion u(x) at location x inside a detected image patch can be well approximated by the following affine motion: u(x) A (x Gamma xm ) u 0 ; 3) with xm the center of the patch, u 0 the displacement of xm , and A a rotation and scaling matrix. Since we ....

Q. Zheng, R. Chellappa, Automatic Feature Point Extraction and Tracking in Image Sequences for Unknown Camera Motion, in Proc. Int. Conf. on Computer Vision, Berlin, Germany, May. 11-14, 1993, pp. 335--339.

....are that provide adequate information for a recognition system to be efficient. One such feature based approach, utilizing a bank of dilated Gabor wavelet filters, has been used in various image understanding recognition algorithms. Applications of these algorithms range from image registration [16] to motion analysis [Wu Chellappa] to a face identification system [10] Intuitively it can be argued that for any system performing face recognition, the first step would be locating the face in the image. Finding the face gives the recognition system a specific area in the image to work on. ....

Q. Zheng and R. Chellappa. Automatic feature point extraction and tracking in image sequences for arbitrary camera motion. Technical Report CAR-TR-628, Center for Automation Research, University of Maryland, College Park, MD, 1992.

....projection of velocities of 3D surface points onto the imaging plane of a visual sensor. The importance of motion in visual processing cannot be understated: in particular, approximations to image motion may be used to estimate 3D scene properties and motion parameters from a moving visual sensor [21, 30, 31, 42, 51, 50, 1, 5, 38, 22, 54, 56, 34, 20, 16, 23], to perform motion segmentation [7, 40, 45, 36, 47, 14, 25, 8, 2, 46, 15] to Submitted to the Journal of Mathematical Imaging and Vision 1999 2 compute the focus of expansion and time to collision [44, 41, 48, 24, 49, 9] to perform motioncompensated image encoding [10, 13, 35, 37, 39, 55] ....

Q. Zheng and R. Chellappa. Automatic feature point extraction and tracking in image sequences for unknown image motion. In Proceedings of ICCV, pages 335--339, Berlin, Germany, May 1993.

....projection of velocities of 3D surface points onto the imaging plane of a visual sensor. The importance of motion in visual processing cannot be understated: in particular, approximations to image motion may be used to estimate 3D scene properties and motion parameters from a moving visual sensor [21, 30, 31, 42, 51, 50, 1, 5, 38, 22, 54, 56, 34, 20, 16, 23], to perform motion segmentation [7, 40, 45, 36, 47, 14, 25, 8, 2, 46, 15] to compute the focus of expansion and time to collision [44, 41, 48, 24, 49, 9] to perform motion compensated image encoding [10, 13, 35, 37, 39, 55] to compute stereo disparity [3, 12, 26, 28] to measure blood flow and ....

Q. Zheng and R. Chellappa. Automatic feature point extraction and tracking in image sequences for unknown image motion. In Proceedings of ICCV, pages 335--339, Berlin, Germany, May 1993. Steven S. Beauchemin John L. Barron

....the greatest probability is retained. 3 Experimental results For the PUMA and Toycar sequences, corners were automatically extracted from each image frame using a variant of the Lucas and Kanade [11] corner detector. 1 For the J7 sequence, we used the corners extracted by Zheng and Chellappa [20]. The MHT parameter values were fixed for all three sequences. Each corner feature was tracked in the image plane using a simple linear Kalman filter with state vector x = x x y y] 0 , where x and y are the pixel coordinates of a feature. The state transition matrix, F, described a constant ....

....is getting closer to the van while for the remainder, the van is receding. This results in tracks whose direction reverse. The relatively large process noise allows the Kalman filter to cope with the change in direction of the tracks. The tracked corners were extracted by Zheng and Chellappa [20]. Comparison with their results show very few differences. Of the 100 tracks, 74 tracks were identical, 18 tracks that contained only one measurement were classified as false alarms by the MHT, 5 tracks with only 3 measurements in each track differed to some degree but these tracks are not ....

[Article contains additional citation context not shown here]

Q. Zheng and R. Chellappa. Automatic feature point extraction and tracking in image sequences for arbitrary camera motion. Int. J. of Computer Vision, (To be published).

....that incorporated geometric constraints and perhaps included a low level perceptual grouping strategy that can identify and group features originating from a common rigid object [9, 12] Reliably detecting corners was also difficult. A coupled feature detection and tracking mechanism, e.g. [19, 20], should be investigated further. The corners were tracked using simple linear Kalman filters. Tuning the various parameters, e.g. process and measurement noise, was straightforward once a few tracks had been manually tracked for several frames. The Puma and J7 sequences demonstrated that (with ....

Q. Zheng and R. Chellappa. Automatic feature point extraction and tracking in image sequences from unknown camera motion. In Fourth Int. Conf. on Computer Vision, pages 335--339, 1993.

....projection of velocities of 3D surface points onto the imaging plane of a visual sensor. The importance of motion in visual processing cannot be understated: in particular, approximations to image motion may be used to estimate 3D scene properties and motion parameters from a moving visual sensor [21, 30, 31, 42, 51, 50, 1, 5, 38, 22, 53, 55, 34, 20, 17, 23], to perform motion segmentation [9, 40, 45, 36, 47, 15, 25, 10, 2, 46, 16] to compute the focus of expansion and time to collision [44, 41, 48, 24, 49, 11] to perform motioncompensated image encoding [12, 14, 35, 37, 39, 54] to compute stereo disparity [3, 13, 26, 28] to measure blood flow ....

Q. Zheng and R. Chellappa. Automatic feature point extraction and tracking in image sequences for unknown image motion. In Proceedings of ICCV, pages 335--339, Berlin, Germany, May 1993.

No context found.

Zheng Q, Chellappa R (1995) Automatic Feature Point Extraction and Tracking in Image Sequences for Arbitrary Camera Motion. Intl Journal of Computer Vision, 15:31--76

No context found.

Zheng Q, Chellappa R (1995) Automatic Feature Point Extraction and Tracking in Image Sequences for Arbitrary Camera Motion. Intl Journal of Computer Vision, 15:31--76

....the number of features to be tracked. Each zone is searched from top to bottom, and the topmost feature is selected for tracking. After the selection of features in frame f t Gamma1 , they can be tracked to frame f t by a multi resolution refinement scheme using parameter estimation proposed in [14]. Although this process is able to produce very good estimates of the motion parameters, it is computationally expensive since it uses a weighted correlation scheme to determine the best feature matches. We use another similarity measure which is given by the SSD over local windows (SSD windows) ....

....can handle can be very large even for small values of s. 2.1. 3 Subpixel Matching After the grid to grid matches are obtained from the hierarchical search, displacements with subpixel accuracy can be easily computed for the finest resolution level of the pyramid using a differential method [10, 14]. Subpixel accuracy is necessary to eliminate the quantization error introduced when the images are digitized. If a feature P 0 t (u; v) has offset (ffix; ffiy) relative to P 0 t Gamma1 (u; v) assume they were tracked and registered so that the translation (ffix; ffiy) is very small) i.e. ....

Q. Zheng and R. Chellappa. Automatic Feature Point Extraction and Tracking in Image Sequences for Unknown Camera Motion. International Journal of Computer Vision, 15:31-76, May 1995.

No context found.

Q. Zheng and R. Chellappa, "Automatic feature point extraction and tracking in image sequences for arbitrary camera motion," Int. J. Comput. Vis., vol. 15, pp. 31--76, 1995.

No context found.

Q. Zheng, R. Chellapa, Automatic feature point extraction and tracking in image sequences for arbitrary camera motion, International Journal of Computer Vision 15 (1995) 31--76.

No context found.

Q. Zheng, R. Chellappa, Automatic Feature Point Extraction and Tracking in Image Sequences for Unknown Camera Motion, in Proc. Int. Conf. on Computer Vision, Berlin, Germany, May. 11-14, 1993, pp. 335--339. 26

No context found.

Q.Zheng and R.Chellappa. Automatic feature point extraction and tracking in image sequences for unknow camera motion. In Fourth International Conference on Computer Vision, 1993.

No context found.

Q. Zheng, R. Chellappa, Automatic Feature Point Extraction and Tracking in Image Sequences for Unknown Camera Motion, in Proc. Int. Conf. on Computer Vision, Berlin, Germany, May. 11-14, 1993, pp. 335--339. This article was processed using the L a T E X macro package with LLNCS style

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