S. Soatto and P. Perona. Recursive estimation of camera motion from uncalibrated image sequences. In Proc. 1 IEEE Int. Conf. on Image Processing, pages III--58--62, Austin -- TX, November 1994. Extended version in: Technical Report CIT-CDS 94-003, California Institute of Technology.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Filter The sensor model (60) is a nonlinear observable system. The extended Kalman filter (EKF) is a widely used scheme to estimate the states of such systems. In the computer vision community, estimation schemes based on Kalman filter have been commonly used for dynamical estimation of motion [23, 25] or road curvature [3, 4] etc. Here, we use the EKF algorithm to estimate on line the 1 ; 2 ; 3 ; and j. Alternatives to the EKF, which are based on nonlinear filtering, are quite complicated and are rarely used. 25 5.2.1 Multiple Measurement Sensor Model In order to make the EKF ....

S. Soatto and P. Perona. Recursive estimation of camera motion from uncalibrated image sequences. In Proceedings ICIP-94, volume 3, pages 58--62, Nov. 1994.

....estimate the motion parameters. Rather, they used it as a smoothing operator on the parameters of motion. At each time step, they computed the difference pose, determined by the essential constraint and expressed by an essential matrix, to update the estimated pose. They extended this idea [Soatto94b] to account for the intrinsic parameters, implying use of the fundamental matrix. The difference fundamental matrix is fundamental by construction, thus the structure required need not be explicitly enforced. They had previously applied the same approach of measuring a difference pose using ....

Soatto, S. and Perona, P. (1994b). Recursive Estimation of Camera Motion from Uncalibrated Image Sequences.InProceedings of the 1 st IEEE International Conference on Image Processing, Pages III--58--62.

....Filter The sensor model (60) is a nonlinear observable system. The extended Kalman filter (EKF) is a widely used scheme to estimate the states of such systems. In the computer vision community, estimation schemes based on Kalman filter have been commonly used for dynamical estimation of motion [23, 25] or road curvature [3, 4] etc. Here, we use the EKF algorithm to estimate on line the 1 ; 2 ; 3 ; and j. Alternatives to the EKF, which are based on nonlinear filtering, are quite complicated and are rarely used. 5.2.1 Multiple Measurement Sensor Model In order to make the EKF ....

S. Soatto and P. Perona. Recursive estimation of camera motion from uncalibrated image sequences. In Proceedings ICIP-94, volume 3, pages 58--62, Nov. 1994.

....in terms of the camera model in question. The simplest cases assume either parallel projection [58, 73, 74, 75] or ideal perspective projection (pinhole model, see [23] More articulated camera models in terms of projective transformations allow parallel and perspective projection as a subcase [3, 25, 61, 70]. We will be mainly concerned with the classical pinhole model; however, our schemes generalize to other camera representations and may estimate the camera model along with visual motion (camera self calibration, see [25, 61] Other schemes recover projective, non metric structure and motion ....

.... allow parallel and perspective projection as a subcase [3, 25, 61, 70] We will be mainly concerned with the classical pinhole model; however, our schemes generalize to other camera representations and may estimate the camera model along with visual motion (camera self calibration, see [25, 61]) Other schemes recover projective, non metric structure and motion independent on the camera parameters [22, 54, 58] Motion reconstruction methods may be further classified in terms of the data processing technique as 2 frames schemes (see for example [38, 49, 78] multiframe batch methods ....

[Article contains additional citation context not shown here]

S. Soatto and P. Perona. Recursive estimation of camera motion from uncalibrated image sequences. In Proc. 1 IEEE Int. Conf. on Image Processing, pages III--58--62, Austin -- TX, November 1994. Extended version in: Technical Report CIT-CDS 94-003, California Institute of Technology.

....the estimate at each step determines a matrix which is fundamental by construction, and we do not need to enforce the structure by solving poorely conditioned polynomial equations. The structure of resulting update is very similar to the essential filter; for details on the derivation see [15]: 2 6 6 4 T R 3 7 7 5 (t 1) 2 6 6 4 T R 3 7 7 5 (t) L(t) 2 6 6 6 4 . x T i (t)A GammaT ( Q( T ; R)A Gamma1 ( x i (t Gamma 1) 3 7 7 7 5 (4) where : fs x ; fs y ; i 0 ; j 0 ] T ; L has the structure of the gain of an ....

....a pixel. Convergence is reached in about 100 frames. Each iteration consists of about 100 Kflops: an implementation using Matlab (not optimized) runs at :6Hz on a Sparc 10 20. We are currently experimenting on real image sequences and higher noise levels. More detailed experiments are reported in [15]. 4 Conclusions We have presented a scheme for estimating ego motion and camera calibration from an image sequence. The scheme is based on an Implicit Extended Kalman Filter in the manifold of fundamental matrices. The update is written in local coordinates, so that at each step the 4 estimated ....

S. Soatto, R. Frezza, and P. Perona. Recursive estimation of camera motion from uncalibrated image sequences. Technical Report CIT-CDS 94-003, California Institute of Technology, 1994.

....in terms of the camera model in question. The simplest cases assume either parallel projection [58, 73, 74, 75] or ideal perspective projection (pinhole model, see [23] More articulated camera models in terms of projective transformations allow parallel and perspective projection as a subcase [3, 25, 61, 70]. We will be mainly concerned with the classical pinhole model; however, our schemes generalize to other camera representations and may estimate the camera model along with visual motion (camera self calibration, see [25, 61] Other schemes recover projective, non metric structure and motion ....

.... allow parallel and perspective projection as a subcase [3, 25, 61, 70] We will be mainly concerned with the classical pinhole model; however, our schemes generalize to other camera representations and may estimate the camera model along with visual motion (camera self calibration, see [25, 61]) Other schemes recover projective, non metric structure and motion independent of the camera parameters [22, 54, 58] Motion reconstruction methods may be further classified in terms of the data processing technique as 2 frames schemes (see for example [38, 49, 78] multiframe batch methods ....

[Article contains additional citation context not shown here]

S. Soatto and P. Perona. Recursive estimation of camera motion from uncalibrated image sequences. In Proc. 1 st IEEE Int. Conf. on Image Processing, pages III--58--62, Austin -- TX, November 1994.

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