## Contour Tracking By Stochastic Propagation of Conditional Density (1996)

Citations: | 593 - 23 self |

### BibTeX

@INPROCEEDINGS{Isard96contourtracking,

author = {Michael Isard and Andrew Blake},

title = {Contour Tracking By Stochastic Propagation of Conditional Density},

booktitle = {},

year = {1996},

pages = {343--356}

}

### Years of Citing Articles

### OpenURL

### Abstract

. In Proc. European Conf. Computer Vision, 1996, pp. 343--356, Cambridge, UK The problem of tracking curves in dense visual clutter is a challenging one. Trackers based on Kalman filters are of limited use; because they are based on Gaussian densities which are unimodal, they cannot represent simultaneous alternative hypotheses. Extensions to the Kalman filter to handle multiple data associations work satisfactorily in the simple case of point targets, but do not extend naturally to continuous curves. A new, stochastic algorithm is proposed here, the Condensation algorithm --- Conditional Density Propagation over time. It uses `factored sampling', a method previously applied to interpretation of static images, in which the distribution of possible interpretations is represented by a randomly generated set of representatives. The Condensation algorithm combines factored sampling with learned dynamical models to propagate an entire probability distribution for object pos...

### Citations

4055 |
Stochastic Relaxation, Gibbs Distributions, and the Bayesian Restoration of Images
- Geman, Geman
- 1984
(Show Context)
Citation Context ... p(zjx) is multi-modal p(xjz) cannot be evaluated simply in closed form: instead iterative sampling techniques can be used. The first use of such an iterative solution was proposed by Geman and Geman =-=[11]-=- for restoration of an image represented by mixed variables, both continuous (pixels) and discrete (the `line process'). Sampling methods for recovery of a parametric curve x by sampling [24, 14, 25] ... |

3277 | Snakes: Active contour models
- Kass, Witkin, et al.
- 1988
(Show Context)
Citation Context ...fficient computationally [20], especially if motion is modelled as well as shape [12, 16]. One important facility is the modelling of curve segments which interact with images [29] or image sequences =-=[19]-=-. This is more general than modelling entire objects but more clutter-resistant than applying signal-processing to low-level corners or edges. The methods to be discussed here have been applied at thi... |

2696 |
R.: Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography
- Fischler, Bolles
- 1981
(Show Context)
Citation Context ...atio-temporal estimatesx(t). With simple, discrete features such as points or corners combinatorial data-association methods can be effective, including the `JPDAF' [2, 22] and the `RANSAC' algorithm =-=[10]-=-. They allow several hypotheses about which data-elements `belong' to the tracked object to be held simultaneously, and less plausible hypotheses to be progressively pruned. Data association methods d... |

779 |
Tracking and Data Association
- Bar-Shalom, Fortmann
- 1988
(Show Context)
Citation Context ...ter which easily `distracts' the spatio-temporal estimatesx(t). With simple, discrete features such as points or corners combinatorial data-association methods can be effective, including the `JPDAF' =-=[2, 22]-=- and the `RANSAC' algorithm [10]. They allow several hypotheses about which data-elements `belong' to the tracked object to be held simultaneously, and less plausible hypotheses to be progressively pr... |

431 |
Recognition by linear combinations of models
- ULLMAN, BASRI
- 1991
(Show Context)
Citation Context ...ir control points. In practice this allows too many degrees of freedom for stable tracking and it is necessary to restrict the curve to a low-dimensional parameter x, for example over an affine space =-=[28, 5]-=-, or more generally allowing a linear space of non-rigid motion [9]. Finally, probability densities p(x) can be defined over the class of curves [9], and also over their motions [27, 5], and this cons... |

363 |
Dynamic 3D models with local and global deformations: deformable superquadrics
- Terzopoulos, Metaxas
- 1991
(Show Context)
Citation Context ...n affine space [28, 5], or more generally allowing a linear space of non-rigid motion [9]. Finally, probability densities p(x) can be defined over the class of curves [9], and also over their motions =-=[27, 5]-=-, and this constitutes a powerful facility for tracking. Reasonable default functions can be chosen for those densities. However, it is obviously more satisfactory to measure the actual densities or e... |

285 |
An Introduction to Splines for Use in Computer Graphics and Geometric Modeling
- Bartels, Beatty, et al.
- 1987
(Show Context)
Citation Context ...ts but more clutter-resistant than applying signal-processing to low-level corners or edges. The methods to be discussed here have been applied at this level, to segments of parametric Bspline curves =-=[3]-=- tracking over image sequences [8]. The B-spline curves could, in theory, be parameterised by their control points. In practice this allows too many degrees of freedom for stable tracking and it is ne... |

246 |
Adaptive Filtering Prediction and Control
- Goodwin, Sin
- 1984
(Show Context)
Citation Context ...ore satisfactory to measure the actual densities or estimate them from data-sequences (x 1 ; x 2 ; : : :). Algorithms to do this assuming Gaussian densities are known in the control-theory literature =-=[13]-=- and have been applied in computer vision [6, 7, 4]. 1.2 Sampling methods A standard problem in statistical pattern recognition is to find an object parameterised as x with prior p(x), using data z fr... |

242 |
Model-based vision: a program to see a walking person
- Hogg
- 1983
(Show Context)
Citation Context ...ision for modelling shape and motion. When suitable geometric models of a moving object are available, they can be matched effectively to image data, though usually at considerable computational cost =-=[17, 26, 18]-=-. Once an object has been located approximately, tracking it in subsequent images becomes more efficient computationally [20], especially if motion is modelled as well as shape [12, 16]. One important... |

213 |
Representations of knowledge in complex systems (with discussion
- Grenander, Miller
- 1994
(Show Context)
Citation Context ...linear estimator is a very special case, applying only to Gaussian densities, of a more general probability density propagation process. In continuous time this can be described in terms of diffusion =-=[15]-=-, governed by a `Fokker-Planck' equation [1], in which the density for x(t) drifts and spreads under the action of a stochastic model of its dynamics. The random component of the dynamical model leads... |

204 |
A Framework for Spatio-Temporal Control in the Tracking of Visual Contours
- Blake, Curwen, et al.
- 1993
(Show Context)
Citation Context ...ir control points. In practice this allows too many degrees of freedom for stable tracking and it is necessary to restrict the curve to a low-dimensional parameter x, for example over an affine space =-=[28, 5]-=-, or more generally allowing a linear space of non-rigid motion [9]. Finally, probability densities p(x) can be defined over the class of curves [9], and also over their motions [27, 5], and this cons... |

172 | model-based motion tracking through the integration of search and estimation
- Lowe
- 1992
(Show Context)
Citation Context ...y to image data, though usually at considerable computational cost [17, 26, 18]. Once an object has been located approximately, tracking it in subsequent images becomes more efficient computationally =-=[20]-=-, especially if motion is modelled as well as shape [12, 16]. One important facility is the modelling of curve segments which interact with images [29] or image sequences [19]. This is more general th... |

169 | Visual tracking of high dof articulated structures: An application to human hand tracking
- Rehg, Kanade
- 1994
(Show Context)
Citation Context ...ation, the tracking of shape and position over time, has been dealt with thoroughly by Kalman filtering, in the relatively clutter-free case in which p(zjx) can satisfactorily be modelled as Gaussian =-=[16, 12, 23]-=- and can be applied to curves [27, 5]. These solutions work relatively poorly in clutter which easily `distracts' the spatio-temporal estimatesx(t). With simple, discrete features such as points or co... |

148 |
Introduction to stochastic control theory
- Åström
- 1970
(Show Context)
Citation Context ...ying only to Gaussian densities, of a more general probability density propagation process. In continuous time this can be described in terms of diffusion [15], governed by a `Fokker-Planck' equation =-=[1]-=-, in which the density for x(t) drifts and spreads under the action of a stochastic model of its dynamics. The random component of the dynamical model leads to spreading --- increasing uncertainty 1 N... |

135 | Tracking known three-dimensional objects
- Gennery
- 1982
(Show Context)
Citation Context ...ional cost [17, 26, 18]. Once an object has been located approximately, tracking it in subsequent images becomes more efficient computationally [20], especially if motion is modelled as well as shape =-=[12, 16]-=-. One important facility is the modelling of curve segments which interact with images [29] or image sequences [19]. This is more general than modelling entire objects but more clutter-resistant than ... |

100 | Learning to track the visual motion of contours
- Blake, Isard, et al.
(Show Context)
Citation Context ...es or estimate them from data-sequences (x 1 ; x 2 ; : : :). Algorithms to do this assuming Gaussian densities are known in the control-theory literature [13] and have been applied in computer vision =-=[6, 7, 4]-=-. 1.2 Sampling methods A standard problem in statistical pattern recognition is to find an object parameterised as x with prior p(x), using data z from a single image. (This is a simplified, static fo... |

96 |
Tracking non-rigid objects in complex scenes
- Huttenlocher, Noh, et al.
- 1992
(Show Context)
Citation Context ...ision for modelling shape and motion. When suitable geometric models of a moving object are available, they can be matched effectively to image data, though usually at considerable computational cost =-=[17, 26, 18]-=-. Once an object has been located approximately, tracking it in subsequent images becomes more efficient computationally [20], especially if motion is modelled as well as shape [12, 16]. One important... |

95 |
Tracking with rigid models
- Harris
- 1992
(Show Context)
Citation Context ...ional cost [17, 26, 18]. Once an object has been located approximately, tracking it in subsequent images becomes more efficient computationally [20], especially if motion is modelled as well as shape =-=[12, 16]-=-. One important facility is the modelling of curve segments which interact with images [29] or image sequences [19]. This is more general than modelling entire objects but more clutter-resistant than ... |

68 | 3D position, attitude and shape input using video tracking of hands and lips”, Computer Graphics (Proc
- Blake, Isard
- 1994
(Show Context)
Citation Context ...es or estimate them from data-sequences (x 1 ; x 2 ; : : :). Algorithms to do this assuming Gaussian densities are known in the control-theory literature [13] and have been applied in computer vision =-=[6, 7, 4]-=-. 1.2 Sampling methods A standard problem in statistical pattern recognition is to find an object parameterised as x with prior p(x), using data z from a single image. (This is a simplified, static fo... |

60 | Interpretation of Known Objects in Constrained Scenes
- Sullivan
- 1992
(Show Context)
Citation Context ...ision for modelling shape and motion. When suitable geometric models of a moving object are available, they can be matched effectively to image data, though usually at considerable computational cost =-=[17, 26, 18]-=-. Once an object has been located approximately, tracking it in subsequent images becomes more efficient computationally [20], especially if motion is modelled as well as shape [12, 16]. One important... |

58 |
The Dynamic Analysis of Apparent Contours
- Cipolla, Blake
- 1990
(Show Context)
Citation Context ... applying signal-processing to low-level corners or edges. The methods to be discussed here have been applied at this level, to segments of parametric Bspline curves [3] tracking over image sequences =-=[8]-=-. The B-spline curves could, in theory, be parameterised by their control points. In practice this allows too many degrees of freedom for stable tracking and it is necessary to restrict the curve to a... |

57 | A bayesian approach to dynamic contours through stochastic sampling and simulated annealing
- Storvik
- 1994
(Show Context)
Citation Context ...nd Geman [11] for restoration of an image represented by mixed variables, both continuous (pixels) and discrete (the `line process'). Sampling methods for recovery of a parametric curve x by sampling =-=[24, 14, 25]-=- have generally used spatial Markov processes as the underlying probabilistic model p(x). The basic method is factored sampling [14]. It is useful when the conditional observation probability p(zjx) c... |

54 |
Building and using flexible models incorporating grey-level information
- Cootes, Taylor, et al.
- 1993
(Show Context)
Citation Context ...for stable tracking and it is necessary to restrict the curve to a low-dimensional parameter x, for example over an affine space [28, 5], or more generally allowing a linear space of non-rigid motion =-=[9]-=-. Finally, probability densities p(x) can be defined over the class of curves [9], and also over their motions [27, 5], and this constitutes a powerful facility for tracking. Reasonable default functi... |

42 |
Deformable templates
- Yuille, Hallinan
- 1992
(Show Context)
Citation Context ...nt images becomes more efficient computationally [20], especially if motion is modelled as well as shape [12, 16]. One important facility is the modelling of curve segments which interact with images =-=[29]-=- or image sequences [19]. This is more general than modelling entire objects but more clutter-resistant than applying signal-processing to low-level corners or edges. The methods to be discussed here ... |

41 | Generating spatiotemporal models from examples
- Baumberg, Hogg
- 1996
(Show Context)
Citation Context ...es or estimate them from data-sequences (x 1 ; x 2 ; : : :). Algorithms to do this assuming Gaussian densities are known in the control-theory literature [13] and have been applied in computer vision =-=[6, 7, 4]-=-. 1.2 Sampling methods A standard problem in statistical pattern recognition is to find an object parameterised as x with prior p(x), using data z from a single image. (This is a simplified, static fo... |

30 |
A Pattern Theoretical Study of Biological Shapes
- HANDS
- 1991
(Show Context)
Citation Context ...nd Geman [11] for restoration of an image represented by mixed variables, both continuous (pixels) and discrete (the `line process'). Sampling methods for recovery of a parametric curve x by sampling =-=[24, 14, 25]-=- have generally used spatial Markov processes as the underlying probabilistic model p(x). The basic method is factored sampling [14]. It is useful when the conditional observation probability p(zjx) c... |

16 |
Finding spiral structures in images of galaxies
- Ripley, Sutherland
- 1990
(Show Context)
Citation Context ...nd Geman [11] for restoration of an image represented by mixed variables, both continuous (pixels) and discrete (the `line process'). Sampling methods for recovery of a parametric curve x by sampling =-=[24, 14, 25]-=- have generally used spatial Markov processes as the underlying probabilistic model p(x). The basic method is factored sampling [14]. It is useful when the conditional observation probability p(zjx) c... |

13 |
Data Association Methods for Tracking Systems
- Rao
- 1993
(Show Context)
Citation Context ...ter which easily `distracts' the spatio-temporal estimatesx(t). With simple, discrete features such as points or corners combinatorial data-association methods can be effective, including the `JPDAF' =-=[2, 22]-=- and the `RANSAC' algorithm [10]. They allow several hypotheses about which data-elements `belong' to the tracked object to be held simultaneously, and less plausible hypotheses to be progressively pr... |