## A Rao-Blackwellized Particle Filter for EigenTracking (2004)

### Cached

### Download Links

Venue: | In IEEE Computer Vision and Pattern Recognition |

Citations: | 67 - 6 self |

### BibTeX

@INPROCEEDINGS{Khan04arao-blackwellized,

author = {Zia Khan and Tucker Balch and Frank Dellaert},

title = {A Rao-Blackwellized Particle Filter for EigenTracking},

booktitle = {In IEEE Computer Vision and Pattern Recognition},

year = {2004},

pages = {980--986}

}

### Years of Citing Articles

### OpenURL

### Abstract

Subspace representations have been a popular way to model appearance in computer vision. In Jepson and Black's influential paper on EigenTracking, they were successfully applied in tracking. For noisy targets, optimization-based algorithms (including EigenTracking) often fail catastrophically after losing track. Particle filters have recently emerged as a robust method for tracking in the presence of multi-modal distributions. To use subspace representations in a particle filter, the number of samples increases exponentially as the state vector includes the subspace coefficients. We introduce an efficient method for using subspace representations in a particle filter by applying Rao-Blackwellization to integrate out the subspace coefficients in the state vector. Fewer samples are needed since part of the posterior over the state vector is analytically calculated. We use probabilistic principal component analysis to obtain analytically tractable integrals. We show experimental results in a scenario in which we track a target in clutter. 1.

### Citations

1502 | Eigenfaces vs. fisherface: recognition using class special linear projection
- Belhumeur, Hespanha, et al.
- 1997
(Show Context)
Citation Context ...tic principal component analysis to obtain analytically tractable integrals. We show experimental results in a scenario in which we track a target in clutter. 1. Introduction Subspace representations =-=[21, 2, 6]-=- have been a longstanding and popular way to model appearance and shape in computer vision. These methods model the density over a high-dimensional space of feature vectors using a generative model, w... |

1496 | Active appearance models
- Cootes, Edwards, et al.
- 2001
(Show Context)
Citation Context ...tic principal component analysis to obtain analytically tractable integrals. We show experimental results in a scenario in which we track a target in clutter. 1. Introduction Subspace representations =-=[21, 2, 6]-=- have been a longstanding and popular way to model appearance and shape in computer vision. These methods model the density over a high-dimensional space of feature vectors using a generative model, w... |

1147 | A tutorial on particle filters for online nonlinear/non-gaussian bayesian tracking
- Arulampalam, Maskell, et al.
(Show Context)
Citation Context ...approximation {X (j) t ,π (j) t } N j=1 for the posterior P (Xt|Z t ) at time t. Note that there are alternative ways to view the particle filter that more easily accommodate other variants (see e.g.=-= [1]),-=- but the mixture proposal view above is well suited for our application. 3. Rao-Blackwellization When modeling a target’s appearance with a subspace model, the state space is augmented with appearan... |

1131 | CONDENSATION conditional density propagation for visual tracking
- Isard, Blake
- 1998
(Show Context)
Citation Context ...ilter framework is to augment the state space of the target with the subspace coefficients. This is the method employed by Blake and Isard to track the an articulated object such as a hand in clutter =-=[12]-=-. However, this proves problematic as the number of samples in the particle filter needs to increase exponentially with the dimensionality of the state space, which now includes the subspace coefficie... |

1058 |
Novel approach to nonlinear/non-Gaussian Bayesian state estimation
- Gordon, Salmond, et al.
- 1993
(Show Context)
Citation Context ...il catastrophically after losing track. Hence, particle filters have recently emerged as a simple and robust method for tracking in the presence of substantial non-normal measurements and/or dynamics =-=[10, 11, 4]-=-. The particle filter approximates the distribution over the current target state as a set of weighted samples, which is recursively updated based on the current measurement, a target motion model, an... |

941 |
Face recognition using eigenface
- Turk, Petland
- 1991
(Show Context)
Citation Context ...tic principal component analysis to obtain analytically tractable integrals. We show experimental results in a scenario in which we track a target in clutter. 1. Introduction Subspace representations =-=[21, 2, 6]-=- have been a longstanding and popular way to model appearance and shape in computer vision. These methods model the density over a high-dimensional space of feature vectors using a generative model, w... |

599 | Detecting Faces in Images: A Survey
- Yang, Kriegman, et al.
- 2002
(Show Context)
Citation Context ...rincipal components analysis (PCA) to a large training set. This approach has been used extensively in a variety of settings, including modeling appearance in face detection and recognition (see e.g. =-=[14] f-=-or a review), modeling 2D shape and appearance [7, 6], and 3D shape and appearance [22]. Subspace representations were also successfully used for tracking, e.g. in Jepson and Black’s influential pap... |

562 | Contour tracking by stochastic propagation of conditional density," ed
- Isard, Blake
- 1996
(Show Context)
Citation Context ...il catastrophically after losing track. Hence, particle filters have recently emerged as a simple and robust method for tracking in the presence of substantial non-normal measurements and/or dynamics =-=[10, 11, 4]-=-. The particle filter approximates the distribution over the current target state as a set of weighted samples, which is recursively updated based on the current measurement, a target motion model, an... |

540 | Eigentracking: Robust matching and tracking of articulated objects using a view-based representation
- Black, Jepson
- 1998
(Show Context)
Citation Context ...ling 2D shape and appearance [7, 6], and 3D shape and appearance [22]. Subspace representations were also successfully used for tracking, e.g. in Jepson and Black’s influential paper on EigenTrackin=-=g [3]-=- in an optimization-based tracking framework. Similarly, Cootes et al [6, 8] model both appearance and shape in a recursive state estimation framework. Unfortunately, for noisy targets, optimizationba... |

476 | Probabilistic principal component analysis
- Tipping, Bishop
- 1999
(Show Context)
Citation Context ...approximated using a more expensive sample representation. We use probabilistic principal component analysis (PPCA), a probabilistic subspace model for which the integral can be computed analytically =-=[20, 17]-=-. Rao-Blackwellized particle filters have been applied in other state estimation problems. In [18], the authors integrate over the 2D target positions and sample over measurement target assignments to... |

448 | FastSLAM: A Factored Solution to the Simultaneous Localization
- Montemerlo, Thrun, et al.
- 2002
(Show Context)
Citation Context ...], de Freitas uses a Rao-Blackwellized particle filter for fault detection where Kalman filters are applied over continuous parameters and samples obtained over discrete fault states. And finally, in =-=[15-=-], the authors integrate over landmark locations in a robotics application where the goal is to localize a robot while simultaneously building a map of the environment. In the remainder of this paper ... |

292 | Use of active shape models for locating structures in medical images
- Cootes, Hill, et al.
- 1994
(Show Context)
Citation Context ...g set. This approach has been used extensively in a variety of settings, including modeling appearance in face detection and recognition (see e.g. [14] for a review), modeling 2D shape and appearance =-=[7, 6], -=-and 3D shape and appearance [22]. Subspace representations were also successfully used for tracking, e.g. in Jepson and Black’s influential paper on EigenTracking [3] in an optimization-based tracki... |

157 | Improved particle filter for nonlinear problems
- Carpenter, Clifford, et al.
- 1999
(Show Context)
Citation Context ...il catastrophically after losing track. Hence, particle filters have recently emerged as a simple and robust method for tracking in the presence of substantial non-normal measurements and/or dynamics =-=[10, 11, 4]-=-. The particle filter approximates the distribution over the current target state as a set of weighted samples, which is recursively updated based on the current measurement, a target motion model, an... |

150 |
Rao-bleickwellisation of Sampling Schemes
- Casella, Robert
- 1996
(Show Context)
Citation Context ... the subspace coefficients [13]. In this paper, we introduce an efficient method for using subspace representations as part of a particle filter. In particular, we propose to use Rao-Blackwellization =-=[16, 5]-=- to integrate out the appearance subspace coefficients of the state vector, leaving only the original target state. The advantage of this is that fewer samples are needed since part of the posterior o... |

98 | Face recognition using active appearance models
- Edwards, Cootes, et al.
- 1998
(Show Context)
Citation Context ...bspace representations were also successfully used for tracking, e.g. in Jepson and Black’s influential paper on EigenTracking [3] in an optimization-based tracking framework. Similarly, Cootes et a=-=l [6, 8]-=- model both appearance and shape in a recursive state estimation framework. Unfortunately, for noisy targets, optimizationbased tracking algorithms often fail catastrophically after losing track. Henc... |

91 |
algorithms for PCA and SPCA
- EM
- 1997
(Show Context)
Citation Context ...approximated using a more expensive sample representation. We use probabilistic principal component analysis (PPCA), a probabilistic subspace model for which the integral can be computed analytically =-=[20, 17]-=-. Rao-Blackwellized particle filters have been applied in other state estimation problems. In [18], the authors integrate over the 2D target positions and sample over measurement target assignments to... |

77 |
Rao-Blackwellised particle filtering for dynamic Bayesian networks
- Murphy, Russell
- 2001
(Show Context)
Citation Context ... the subspace coefficients [13]. In this paper, we introduce an efficient method for using subspace representations as part of a particle filter. In particular, we propose to use Rao-Blackwellization =-=[16, 5]-=- to integrate out the appearance subspace coefficients of the state vector, leaving only the original target state. The advantage of this is that fewer samples are needed since part of the posterior o... |

74 |
The Wisdom of the Hive
- Seeley
- 1996
(Show Context)
Citation Context ...llized particle filter to track an unmarked honey bee in an observation hive. The automated recording of trajectory data has significant applications in the study of honey bee behavior and physiology =-=[19]-=-. This application presents substantial challenges due to temporary occlusions that occur in the close proximity of other bees, complex variations in the appearance of the tracked bee, and the unpredi... |

55 | People tracking with anonymous and id-sensors using rao-blackwellised particle filters
- Schulz, Fox, et al.
- 2003
(Show Context)
Citation Context ...alysis (PPCA), a probabilistic subspace model for which the integral can be computed analytically [20, 17]. Rao-Blackwellized particle filters have been applied in other state estimation problems. In =-=[18]-=-, the authors integrate over the 2D target positions and sample over measurement target assignments to track people based on noisy position measurements from IR sensors. In [9], de Freitas uses a Rao-... |

30 | Efficient particle filter-based tracking of multiple interacting targets using an mrf-based motion model
- Khan, Balch, et al.
- 2003
(Show Context)
Citation Context ...wever, this proves problematic as the number of samples in the particle filter needs to increase exponentially with the dimensionality of the state space, which now includes the subspace coefficients =-=[13]-=-. In this paper, we introduce an efficient method for using subspace representations as part of a particle filter. In particular, we propose to use Rao-Blackwellization [16, 5] to integrate out the ap... |

30 | Estimating Coloured 3D Face Models from Single Images: An Example Based Approach
- Vetter, Blanz
- 1998
(Show Context)
Citation Context ...xtensively in a variety of settings, including modeling appearance in face detection and recognition (see e.g. [14] for a review), modeling 2D shape and appearance [7, 6], and 3D shape and appearance =-=[22]. -=-Subspace representations were also successfully used for tracking, e.g. in Jepson and Black’s influential paper on EigenTracking [3] in an optimization-based tracking framework. Similarly, Cootes et... |

9 |
Rao-Blackwellized particle filtering for fault diagnosis
- Freitas
- 2002
(Show Context)
Citation Context ...stimation problems. In [18], the authors integrate over the 2D target positions and sample over measurement target assignments to track people based on noisy position measurements from IR sensors. In =-=[9]-=-, de Freitas uses a Rao-Blackwellized particle filter for fault detection where Kalman filters are applied over continuous parameters and samples obtained over discrete fault states. And finally, in [... |