R. Szeliski, "Shape from Rotation," Proc. IEEE Conf. Computer Vision and Pattern Recognition, pp. 625-630, 1991.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....information is to get the 3D location of some surface points and use these points as starting points for our surface recovery procedure. The stereo information can also be used to remove the ambiguity in the q component of surface orientation. The integration of all the cues is not an easy task [8, 9, 6, 18, 1]. Extending our work to more complicated surfaces will require integrating other cues. Another extension of our work is surface recovery by rotating the object more than 90 degrees. In this way, we can get more singular points and obtain a more accurate estimate of the reflectance function. We ....

R. Szeliski. Shape from rotation. In Proc. IEEE Conf. Computer Vision and Pattern Recognition, 1991, pages 625--630, June 1991.

....using techniques such as those of Faugeras [3] Shashua [15] Hartley, 5] and Heyden [6] The structure obtained is in the same projective family as the Euclidean structure, but may be distorted. Euclidean structure from a calibrated camera can be found using algorithms such as those of Szeliski [18] and Matsumoto [11] More recently, techniques have been devised for obtaining Euclidean structure from motion using an uncalibrated camera. Heyden and Astrom [7] obtain structure with constant but unknown camera parameters. They recover an intermediate projective structure to compute an ....

R. Szeliski. Shape from rotation. In IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'91), pages 625--630, Maui, Hawaii, June 1991. IEEE Computer Society Press.

....rather than just the frontal views. The overall goal of our research is to devise an accurate method for reconstructing the complete 3D models of small objects. To ensure reconstruction accuracy, the method uses a calibrated camera and a computer controlled turntable, similar to the setup in [13, 32, 33, 34]. Calibration of the camera s intrinsic and extrinsic parameters can be performed using standard calibration algorithms such as [10, 36] with the calibration object placed on the turntable. Specialized calibration algorithms adapted to rotating objects such as [33] can also be used. The turntable ....

....reconstruction under unknown but complete orbital camera motion. Although the method in [30] can potentially recover complete 3D models, only the reconstructed frontal views are shown in the article. The methods by Matsumoto et al. 18] Mehren and Rodehorst [14, 19] and Szeliski and co workers [13, 32, 33, 34] are the most similar to ours. All these methods attempt to recover Euclidean structures from rotating objects. In particular, the methods of Matsumoto et al. and Mehren and Rodehorst reconstruct object models from image silhouette. Mehren and Rodehorst also introduced another method that recover ....

[Article contains additional citation context not shown here]

R. Szeliski. Shape from rotation. In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pages 625--630, 1991.

.... body of work on the recovery of raw 3 D data from multiple images; they include multibaseline stereo [14] trinocular stereo that combines constant brightness constraint with trilinear tensor (small displacements, only three images) 19] stereo with interpolation [4] and shape from rotation [21, 30]. In a work that unifies image matching with stereo, Xu and Zhang [29] use 2 1 INTRODUCTION initially extracted correspondence to estimate the epipolar geometry using a robust estimator. The computed epipolar geometry is then used to recover more correspondences as in classical stereo matching. ....

R. Szeliski. Shape from rotation. In IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'91), pages 625--630, Maui, Hawaii, June 1991. IEEE Computer Society Press.

.... have used a variety of single image cues for reconstruction, such as silhouettes and image features (Cohen et al. 1991, Delingette et al. 1991, Terzopoulos et al. 1987, Tomasi and Kanade 1992, Wang and Wang 1992) range data (Whaite and Ferrie 1991) stereo (Fua and Sander 1992) and motion (Szeliski 1991). Liedtke et al. 1991) first uses silhouettes to derive an initial estimate of the surface, and then uses a multi image stereo algorithm to improve on the result. Both their approach to deriving an initial estimate for the mesh and Szeliski and Tonnesen s approach (1992) are different from ours ....

Szeliski, R. (1991). Shape from rotation. In Conference on Computer Vision and Pattern Recognition, pages 625--630.

....to use contours to remove this limitation. Stereo information can provide the 3D location of starting points for our surface recovery procedure. The stereo information can also remove the ambiguity in the q component of surface orientation. The integration of all the cues is not an easy task [8, 9, 17, 1]. Extending our work to more complicated surfaces will require integrating other cues. Another extension of our work is surface recovery by rotating the object more than 90 degrees. In this way, we can get more singular points and obtain a more accurate estimate of the reflectance function. We can ....

R. Szeliski. Shape from rotation. In CVPR-91, pages 625--630, June 1991.

....However, the advantage of this approximation is a simple, non iterative estimation algorithm [23] In contrast, specializing the motion to single axis is an exact model of the geometry, not an approximation, yet it admits a closed form solution. Previous investigations of turn table sequences [12, 20, 22] have not fully exploited the special motion to simplify camera recovery. 2 The projective geometry of single axis motion A single axis motion consists of a set of Euclidean actions on the world such that the relative motion between the scene and camera can be described by rotations about a ....

R. Szeliski. Shape from rotation. In Proc. CVPR, pages 625--630, 1991.

....the 3D locations of some surface points from surface features and use these points as starting points for our surface recovery procedure. The stereo information can also be used to remove the ambiguity on the q component of surface orientation. The integration of all the cues is not an easy task [9, 10, 7, 18, 1]. extending our work to surfaces with piecewise uniform reflectance will require integrating different cues. Another extension of our work is surface recovery by rotating the object more than 90 degrees. In this way, we can get more singular points and obtain a more accurate estimate of the ....

R. Szeliski. Shape from rotation. In Proc. IEEE Conf. Computer Vision and Pattern Recognition, 1991, pages 625--630, June 1991.

....calibrating to determine the precise location of the camera. Thus it is natural to consider rotating an object in front of a fixed camera. In photometric stereo, to recover the surface orientation of a whole object, Woodham [10] uses a rotary table to rotate the object. In shape from rotation [5, 6], Szeliski places an object on a spring wound microwave turntable and uses optical flow information or contour information to derive the 3D structure of the object. In 3D model acquisition [11] Zheng puts an object on a turntable or a person on a swivel chair and uses contour information to ....

....reflectance function can be extracted from an image sequence and a relatively accurate initial surface can be estimated. Moreover the interreflection under the collinear light source is easy to analyze. Compared to other techniques of surface recovery from image sequence of a rotating object [12, 11, 5, 6], our technique has certain advantages. First it does not require a continuous image sequence. Secondly it can get a dense map of both surface depth and orientation at the same time. Finally our technique works on any surface of isotropic reflectance function. The intended applications of our ....

R. Szeliski. Shape from rotation. In CVPR-91, pages 625--630, June 1991.

....many techniques for extracting 3 D shape from multiple views. For example, we can recover a volumetric description from the binary silhouettes of an object against its background [Szeliski, 1993] compute local optic flow (pixel motion) estimates and convert these into sparse 3 D point estimates [Szeliski, 1991], or track the occluding contours of an object to generate 3 D space curves [Szeliski and Weiss, 1993] A complete survey of 3 D shape extraction techniques is beyond the scope of this paper. Instead, we present the results of two of our previously developed algorithms applied to an image sequence ....

R. Szeliski. Shape from rotation. In IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'91), pages 625--630, IEEE Computer Society Press, Maui, Hawaii, June 1991. 10 Conclusions 27

....obtained at the expense of fewer recovered points. We have also applied our algorithm to the four real image sequences shown in Figure 6. These sequences were obtained by placing an object on a rotating mechanized turntable whose edge has a Gray code strip used for reading back the rotation angle [Sze91, Sze93]. The camera motion parameters for these sequences were obtained by first calibrating the camera intrinsic parameters and extrinsic parameters to the turntable top center, and then using the computed turntable rotation. Figure 8 shows two views of each set of reconstructed 3D curves. We can see ....

R. Szeliski. Shape from rotation. In IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'91), pages 625--630, Maui, Hawaii, June 1991. IEEE Computer Society Press.

....performing slightly better than linear smoothing. We have also applied our algorithm to the four real image sequences shown in Figure 6. These sequences were obtained by placing an object on a rotating mechanized turntable whose edge has a Gray code strip used for reading back the rotation angle [Szeliski, 1991; Szeliski, 1993] The camera motion parameters for these sequences were obtained by first calibrating the camera intrinsic parameters and extrinsic parameters to the turntable top center, and then using the computed turntable rotation. Figure 7 shows the edges extracted from each of these images. ....

R. Szeliski. Shape from rotation. In IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'91), pages 625--630, IEEE Computer Society Press, Maui, Hawaii, June 1991.

.... is not limited to a particular topology, unlike previous 3 D surface fitting models such as [Terzopoulos et al. 1987b; Miller et al. 1991] We can also fit surfaces to data that does not originate from closed surfaces, such as stereo range data [Barnard and Fischler, 1982; Fua and Sander, 1992; Szeliski, 1991] Simply growing particles away from the sample points poses several problems. For example, if we allow growth in all directions, the surface may grow indefinitely at the edges, whereas if we limit the growth at edges, we may not be able to fill in certain gaps. Instead, we apply the stretching ....

....section, oriented particles provide a solution by extending the surface out from known data points. We believe that these techniques will be particularly useful in machine vision applications where it can be used to interpolate sparse position measurements available from stereo or tactile sensing [Szeliski, 1991]. A direct extension of our surface fitting procedure is to add a potential function that induces a torque around the local z axis. This torque can be used to force the x and y axes to align themselves in the directions of minimum and maximum curvature. For example, the potential term (expressed ....

R. Szeliski. Shape from rotation. In IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'91), pages 625--630, IEEE Computer Society Press, Maui, Hawaii, June 1991.

....many techniques for extracting 3 D shape from multiple views. For example, we can recover a volumetric description from the binary silhouettes of an object against its background [Szeliski, 1993] compute local optic flow (pixel motion) estimates and convert these into sparse 3 D point estimates [Szeliski, 1991], or track the occluding contours of an object to generate 3 D space curves [Szeliski and Weiss, 1993] A complete survey of 3 D shape extraction techniques is beyond the scope of this paper. Instead, we present the results of two of our previously developed algorithms applied to an image sequence ....

R. Szeliski. Shape from rotation. In IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'91), pages 625--630, IEEE Computer Society Press, Maui, Hawaii, June 1991. 10 Conclusions 27

No context found.

R. Szeliski, "Shape from Rotation," Proc. IEEE Conf. Computer Vision and Pattern Recognition, pp. 625-630, 1991.

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