## Kernel-Based Object Tracking (2003)

### Cached

### Download Links

- [www.caip.rutgers.edu]
- [www.wisdom.weizmann.ac.il]
- [www.comaniciu.net]
- [www.soe.ucsc.edu]
- [www.cim.mcgill.ca]
- [www.cim.mcgill.ca]
- [www.wisdom.weizmann.ac.il]
- [classes.soe.ucsc.edu]
- [comaniciu.net]
- [www.xmt.be]
- [www.cs.drexel.edu]
- [web.eecs.umich.edu]
- [www.caip.rutgers.edu]
- [www.caip.rutgers.edu]
- [www.ee.duke.edu]
- [www.cs.ucf.edu]
- DBLP

### Other Repositories/Bibliography

Citations: | 668 - 3 self |

### BibTeX

@MISC{Comaniciu03kernel-basedobject,

author = {Dorin Comaniciu and Visvanathan Ramesh and Peter Meer},

title = {Kernel-Based Object Tracking},

year = {2003}

}

### Years of Citing Articles

### OpenURL

### Abstract

A new approach toward target representation and localization, the central component in visual tracking of non-rigid objects, is proposed. The feature histogram based target representations are regularized by spatial masking with an isotropic kernel. The masking induces spatially-smooth similarity functions suitable for gradient-based optimization, hence, the target localization problem can be formulated using the basin of attraction of the local maxima. We employ a metric derived from the Bhattacharyya coefficient as similarity measure, and use the mean shift procedure to perform the optimization. In the presented tracking examples the new method successfully coped with camera motion, partial occlusions, clutter, and target scale variations. Integration with motion filters and data association techniques is also discussed. We describe only few of the potential applications: exploitation of background information, Kalman tracking using motion models, and face tracking. Keywords: non-rigid object tracking; target localization and representation; spatially-smooth similarity function; Bhattacharyya coefficient; face tracking. 1

### Citations

9590 | Elements of Information Theory - Cover, Thomas - 1991 |

4789 | A tutorial on hidden markov models and selected applications in speeach recognition
- Rabiner
- 1989
(Show Context)
Citation Context ...cretely sampled points to parameterize the mean and covariance of the posterior density. When the state space is discrete and consists of a finite number of states, Hidden Markov Models (HMM) filters =-=[60]-=- can be applied for tracking. The most general class of filters is represented by particle filters [45], also called bootstrap filters [31], which are based on Monte Carlo integration methods. The cur... |

3038 | Introduction to Statistical Pattern Recognition, 2nd ed - Fukunaga - 1990 |

1717 | Mean shift: A robust approach toward feature space analysis
- Comaniciu, Meer
- 2002
(Show Context)
Citation Context ... kðxÞ at y in the current frame, with the data being weighted by wi (10). The mode of this density in the local neighborhood is the sought maximum that can be found employing the mean shift procedur=-=e [17-=-]. In this procedure, the kernel is recursively moved from the current location ^y 0 to the new location ^y 1 according to the relation ^y 1 P nh i1 xiwig P nh i1 wig ^y 0 xi h ^y 0 xi h 2 2 ; ð11Þ ... |

1450 | Color indexing
- Swain
- 1990
(Show Context)
Citation Context .... Therefore, Ch can be precalculated for a given kernel and different values of h. The 2. It has a clear geometric interpretation. Note that the Lp histogram metrics (including histogram intersection =-=-=-[71]) do not enforce the conditions Pm u1 ^qu 1 and Pm u1 ^pu 3. 4. 1. It uses discrete densities and, therefore, it is invariant to the scale of the target (up to quantization effects). It is valid f... |

1426 | A tutorial on particle filters for online nonlinear/non-gaussian bayesian tracking
- Arulampalam, Maskell, et al.
(Show Context)
Citation Context ...o integration methods. The current density of the state is represented by a set of random samples with associated weights and the new density is computed based on these samples and weights (see [23], =-=[3]-=- for reviews). The UKF can be employed to generate proposal distributions for particle filters, in which case the filter is called Unscented Particle Filter (UPF) [54]. When the tracking is performed ... |

1270 | Condensation – conditional density propagation for visual tracking
- Isard, Blake
- 1998
(Show Context)
Citation Context ...laroff [65] used the Extended Kalman Filter to estimate a 3D object trajectory from 2D image motion. Particle filtering was first introduced in vision as the Condensation algorithm by Isard and Blake =-=[40]-=-. Probabilistic exclusion for tracking multiple objects was discussed in [51]. Wu and Huang developed an algorithm to integrate multiple target clues [76]. Li and Chellappa [48] proposed simultaneous ... |

1265 | Pfinder: Real-time tracking of the human body
- Wren, Azarbayehani, et al.
- 1997
(Show Context)
Citation Context ...N REAL-TIME object tracking is the critical task in many computer vision applications such as surveillance [44], [16], [32], perceptual user interfaces [10], augmented reality [26], smart rooms [39], =-=[75]-=-, [47], object-based video compression [11], and driver assistance [34], [4]. Two major components can be distinguished in a typical visual tracker. Target Representation and Localization is mostly a ... |

1263 |
Novel approach to nonlinear/non-Gaussian Bayesian state estimation
- Gordon, Salmond, et al.
- 1993
(Show Context)
Citation Context ... finite number of states, Hidden Markov Models (HMM) filters [60] can be applied for tracking. The most general class of filters is represented by particle filters [45], also called bootstrap filters =-=[31]-=-, which are based on Monte Carlo integration methods. The current density of the state is represented by a set of random samples with associated 2 weights and the new density is computed based on thes... |

844 | III. Alignment by maximization of mutual information
- Viola, Wells
- 1997
(Show Context)
Citation Context ...lization. While the filtering and data association have their roots in control theory, algorithms for target representation and localization are specific to images and related to registration methods =-=[72]-=-, [64], [56]. Both target localization and registration maximizes a likelihood type function. The difference is that in tracking, as opposed to registration, only small changes are assumed in the loca... |

814 |
Tracking and data association
- Bar-Shalon, Fortmann
- 1988
(Show Context)
Citation Context ... complexity of a tracker as low as possible. The most abstract formulation of the filtering and data association process is through the state space approach for modeling discrete-time dynamic systems =-=[5]-=-. The information characterizing the target is defined by the state sequencefxkg k0;1;... , whose evolution in time is specified by the dynamic equation xk f kðxk 1; vkÞ. The available . D. Comanici... |

778 | On sequential Monte Carlo sampling methods for Bayesian
- Doucet, Godsill, et al.
- 2000
(Show Context)
Citation Context ...e Carlo integration methods. The current density of the state is represented by a set of random samples with associated weights and the new density is computed based on these samples and weights (see =-=[23]-=-, [3] for reviews). The UKF can be employed to generate proposal distributions for particle filters, in which case the filter is called Unscented Particle Filter (UPF) [54]. When the tracking is perfo... |

667 | Real-time tracking of non-rigid objects using mean shift
- Comaniciu, Ramesh, et al.
(Show Context)
Citation Context ...e can introduce a large bias in the estimated location of the target, and the resulting measure is scale variant (see [37, p. 262] for a discussion). We mention that since its original publication in =-=[18]-=-, the idea of kernel-based tracking has been exploited and developed forward by various researchers. Chen and Liu [14] experimented with the same kernel-weighted histograms, but employed the Kullback-... |

648 | The visual analysis of human movement: a survey
- Gavrila
- 1999
(Show Context)
Citation Context ...ted an affine tracker based on planar regions and anchor points. Tracking people, which raises many challenges due to the presence of large 3D, nonrigid motion, was extensively analyzed in [36], [1], =-=[30]-=-, [73]. Explicit tracking approaches of people [69] are timeconsuming and often the simpler blob model [75] or adaptive mixture models [53] are also employed. The main contribution of the paper is to ... |

544 | New extension of the kalman filter to nonlinear systems
- Julier, Uhlmann
- 1997
(Show Context)
Citation Context ... by linearization the Extended Kalman Filter (EKF) [5, p. 106] is obtained, the posterior density being still modeled as Gaussian. A recent alternative to the EKF is the Unscented Kalman Filter (UKF) =-=[42]-=- which uses a set of discretely sampled points to parameterize the mean and covariance of the posterior density. When the state space is discrete and consists of a finite number of states, Hidden Mark... |

482 | An algorithm for tracking multiple targets
- Reid
- 1979
(Show Context)
Citation Context ... p. 222], on the other hand, calculates the measurement-to-target association probabilities jointly across all the targets. A different strategy is represented by the Multiple Hypothesis Filter (MHF) =-=[63]-=-, [20], [5, p. 106] which evaluates the probability that a given target gave rise to a certain measurement sequence. The MHF formulation can be adapted to track the modes of the state density [13]. Th... |

476 | Divergence measures based on the shannon entropy
- Lin
- 1991
(Show Context)
Citation Context ...ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ^puðyÞ^qu ; ð7Þ u1 the sample estimate of the Bhattacharyya coefficient between p and q [43]. The Bhattacharyya coefficient is a divergence-type measu=-=re -=-[49] which has a straightforward geometric interpretawhere is the Kronecker delta function. The normalization constant C is derived by imposing the condition Pm u1 ^qu 1, from where 1 C Pn i1 k kx? ik... |

421 | Non-parametric Model for Background Subtraction
- Elgammal, Harwood, et al.
- 2000
(Show Context)
Citation Context ... of attraction of the mode covers the entire rectangular window. In controlled environments with fixed camera, additional geometric constraints (such as the expected scale) and background subtraction =-=[24]-=- can be exploited to improve the tracking process. The Subway-1 sequence (Fig. 4) is suitable for such an approach, however, the results presented here has been processed with the algorithm unchanged.... |

418 | Multivariate Density Estimation
- Scott
- 1992
(Show Context)
Citation Context ...ams should be used. Thus, we have target model : ^q fg ^qu u1...m target candidate : ^pðyÞf^puðyÞgu1...m X m u1 Xm u1 ^qu 1 ^pu 1: The histogram is not the best nonparametric density estimate [68]=-=, but it-=- suffices for our purposes. Other discrete density estimates can be also employed. We will denote by ^ðyÞ ^pðyÞ; ^qŠ ð1Þ a similarity function between ^p and ^q. The function ^ðyÞ plays the r... |

391 |
A.: CONDENSATION-Conditional Density Propagation for Visual Tracking
- Isard, Blake
- 1998
(Show Context)
Citation Context ...roff [65] used the Extended Kalman Filter to estimate a 3D object trajectory from 2D image motion. Particle filtering was first introduced, in vision, as the Condensation algorithm by Isard and Blake =-=[40]-=-. Probabilistic exclusion for tracking multiple objects was discussed in [51]. Wu and Huang developed an algorithm to integrate multiple target clues [76]. Li and Chellappa [48] proposed simultaneous ... |

348 | Human motion analysis: a review
- Aggarwal, Cai
- 1997
(Show Context)
Citation Context ...[26] presented an affine tracker based on planar regions and anchor points. Tracking people, which rises many challenges due to the presence of large 3D, non-rigid motion, was extensively analyzed in =-=[36, 1, 30, 73]-=-. Explicit tracking approaches of people [69] are time-consuming and often the simpler blob model [75] or adaptive mixture models [53] are also employed. The main contribution of the paper is to intro... |

293 | Elliptical head tracking using intensity gradients and color histograms
- Birchfield
- 1998
(Show Context)
Citation Context ...f iterations is 4:19 per frame. region inside the rectangle. The surface is asymmetric due to neighboring colors that are similar to the target. While most of the tracking approaches based on regions =-=[7]-=-, [27], [50] must perform an exhaustive search in the rectangle to find the maximum, our algorithm converged in four iterations as shown in Fig. 3. Note that the operational basin of attraction of the... |

281 | Color-based probabilistic tracking
- Pérez, Hue, et al.
- 2002
(Show Context)
Citation Context ... texture analysis. The benefits of guiding random particles by gradient optimization are discussed in [70] and a particle filter for color histogram tracking based on the metric (6) is implemented in =-=[57]-=-. Finally, we would like to add a word of caution. The tracking solution presented in this paper has several desirable properties: it is efficient, modular, has straightforwardsCOMANICIU ET AL.: KERNE... |

274 | Robust online appearance models for visual tracking
- Jepson, Fleet, et al.
(Show Context)
Citation Context ...ed by Sclaroff and Isidoro [67] using robust M-estimators. Learning of appearance models by employing a mixture of stable image structure, motion information, and an outlier process, was discussed in =-=[41]-=-. In a different approach, Ferrari et al. [26] presented an affine tracker based on planar regions and anchor points. Tracking people, which raises many challenges due to the presence of large 3D, non... |

231 | A probabilistic exclusion principle for tracking multiple objects
- MacCormick, Blake
- 1999
(Show Context)
Citation Context ...from 2D image motion. Particle filtering was first introduced, in vision, as the Condensation algorithm by Isard and Blake [40]. Probabilistic exclusion for tracking multiple objects was discussed in =-=[51]-=-. Wu and Huang developed an algorithm to integrate multiple target clues [76]. Li and Chellappa [48] proposed simultaneous tracking and verification based on particle filters applied to vehicles and f... |

220 | Moving target classification and tracking from real-time video
- Lipton, Fujiyoshi, et al.
- 1998
(Show Context)
Citation Context ...edges, or any combination of them. In the sequel, it is assumed that the following information is available: 1) detection and localization in the initial frame of the objects to track (target models) =-=[50]-=-, [8] and 2) periodic analysis of each object to account for possible updates of the target models due to significant changes in color [53]. 4.1 Distance Minimization Minimizing the distance (6) is eq... |

215 |
The divergence and Bhattacharyya distance measures in signal selection
- Kailath
- 1967
(Show Context)
Citation Context ...ffiffi p 1 ^pðyÞ; ^qŠ; ð6Þ where we chose ^ðyÞ ^pðyÞ; ^qŠ Xm pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ^puðyÞ^qu ; ð7Þ u1 the sample estimate of the Bhattacharyya coefficient betwee=-=n p and q [43]-=-. The Bhattacharyya coefficient is a divergence-type measure [49] which has a straightforward geometric interpretawhere is the Kronecker delta function. The normalization constant C is derived by impo... |

214 | Empirical evaluation of dissimilarity measures for color and texture
- Puzicha, Buhmann, et al.
- 1999
(Show Context)
Citation Context ...ces an interpolation process between the locations on the image lattice. The employed target representations do not restrict the way similarity is measured and various functions can be used for . See =-=[59]-=- for an experimental evaluation of different histogram similarity measures. 3 METRIC BASED oN BHATTACHARYYA COEFFICIENT The similarity function defines a distance among target model and candidates. To... |

196 | A multiple hypothesis approach to figure tracking
- Cham, Rehg
- 1999
(Show Context)
Citation Context ...HF) [63], [20], [5, p. 106] which evaluates the probability that a given target gave rise to a certain measurement sequence. The MHF formulation can be adapted to track the modes of the state density =-=[13]-=-. The data association problem for multiple target particle filtering is presented in [62], [38]. The filtering and association techniques discussed above were applied in computer vision for various t... |

192 | S.: Multi-Camera MultiPerson Tracking for EasyLiving
- Krumm, Harris, et al.
- 2000
(Show Context)
Citation Context ...-TIME object tracking is the critical task in many computer vision applications such as surveillance [44], [16], [32], perceptual user interfaces [10], augmented reality [26], smart rooms [39], [75], =-=[47]-=-, object-based video compression [11], and driver assistance [34], [4]. Two major components can be distinguished in a typical visual tracker. Target Representation and Localization is mostly a bottom... |

183 |
An Efficient Implementation of Reid's Multiple Hypothesis Tracking Algorithm and its Evaluation for the Purpose of Visual Tracking
- Cox, Hingorani
- 1996
(Show Context)
Citation Context ...2], on the other hand, calculates the measurement-to-target association probabilities jointly across all the targets. A different strategy is represented by the Multiple Hypothesis Filter (MHF) [63], =-=[20]-=-, [5, p. 106] which evaluates the probability that a given target gave rise to a certain measurement sequence. The MHF formulation can be adapted to track the modes of the state density [13]. The data... |

182 | Support vector tracking
- Avidan
- 2001
(Show Context)
Citation Context ...ications such as surveillance [44], [16], [32], perceptual user interfaces [10], augmented reality [26], smart rooms [39], [75], [47], object-based video compression [11], and driver assistance [34], =-=[4]-=-. Two major components can be distinguished in a typical visual tracker. Target Representation and Localization is mostly a bottom-up process which has also to cope with the changes in the appearance ... |

179 | Algorithms for cooperative multisensor surveillance
- Collins, Lipton, et al.
- 2001
(Show Context)
Citation Context ...lly-smooth similarity function, Bhattacharyya coefficient, face tracking. 1 INTRODUCTION REAL-TIME object tracking is the critical task in many computer vision applications such as surveillance [44], =-=[16]-=-, [32], perceptual user interfaces [10], augmented reality [26], smart rooms [39], [75], [47], object-based video compression [11], and driver assistance [34], [4]. Two major components can be disting... |

176 |
Non-Gaussian state-space modeling of nonstationary time series: Rejoinder
- Kitagawa
- 1987
(Show Context)
Citation Context ... space is discrete and consists of a finite number of states, Hidden Markov Models (HMM) filters [60] can be applied for tracking. The most general class of filters is represented by particle filters =-=[45]-=-, also called bootstrap filters [31], which are based on Monte Carlo integration methods. The current density of the state is represented by a set of random samples with associated weights and the new... |

176 | w : Who? when? where? what? a real time system for detecting and tracking people
- Haritaoglu, Harwood, et al.
(Show Context)
Citation Context ...[26] presented an affine tracker based on planar regions and anchor points. Tracking people, which rises many challenges due to the presence of large 3D, non-rigid motion, was extensively analyzed in =-=[36, 1, 30, 73]-=-. Explicit tracking approaches of people [69] are time-consuming and often the simpler blob model [75] or adaptive mixture models [53] are also employed. The main contribution of the paper is to intro... |

144 | B.: Covariance scaled sampling for monocular 3d body tracking
- Sminchisescu, Triggs
(Show Context)
Citation Context ...nchor points. Tracking people, which raises many challenges due to the presence of large 3D, nonrigid motion, was extensively analyzed in [36], [1], [30], [73]. Explicit tracking approaches of people =-=[69]-=- are timeconsuming and often the simpler blob model [75] or adaptive mixture models [53] are also employed. The main contribution of the paper is to introduce a new framework for efficient tracking of... |

131 | Probabilistic data association methods for tracking complex visual objects
- Rasmussen, Hager
(Show Context)
Citation Context ...to a certain measurement sequence. The MHF formulation can be adapted to track the modes of the state density [13]. The data association problem for multiple target particle filtering is presented in =-=[62]-=-, [38]. The filtering and association techniques discussed above were applied in computer vision for various tracking scenarios. Boykov and Huttenlocher [9] employed the Kalman filter to track vehicle... |

128 | M.: View-invariant representation and recognition of actions
- Rao, Yilmaz, et al.
(Show Context)
Citation Context ...(FLIR) imagery. Xu and Fujimura [77] used night vision for pedestrian detection and tracking, where the detection is performed by a support vector machine and the tracking is kernel-based. Rao et al. =-=[61]-=- employed kernel tracking in their system for action recognition, while Caenen et al. [12] followed the same principle for texture analysis. The benefits of guiding random particles by gradient optimi... |

123 |
Tracking persons in monocular image sequences
- Wachter, Nagel
- 1999
(Show Context)
Citation Context ... affine tracker based on planar regions and anchor points. Tracking people, which raises many challenges due to the presence of large 3D, nonrigid motion, was extensively analyzed in [36], [1], [30], =-=[73]-=-. Explicit tracking approaches of people [69] are timeconsuming and often the simpler blob model [75] or adaptive mixture models [53] are also employed. The main contribution of the paper is to introd... |

122 | Optical Flow Constraints on Deformable Models with Applications to Face Tracking’, CIS technical report MS-CIS-97-23
- DeCarlo, Metaxas
(Show Context)
Citation Context ...ication dependent and plays a decisive role in the robustness and efficiency of the tracker. For example, face tracking in a crowded scene relies more on target representation than on target dynamics =-=[21]-=-, while in aerial video surveillance, e.g., [74], the target motion and the ego-motion of the camera are the more important components. In real-time applications, only a small percentage of the system... |

110 | Real-time tracking of image regions with changes in geometry and illumination
- Hager, Belhumeur
- 1996
(Show Context)
Citation Context ...erty can be exploited to develop efficient, gradient-based localization schemes using the normalized correlation criterion [6]. Since the correlation is sensitive to illumination, Hager and Belhumeur =-=[33]-=- explicitly modeled the geometry and illumination changes. The method was improved by Sclaroff and Isidoro [67] using robust M-estimators. Learning of appearance models by employing a mixture of stabl... |

107 |
Bayesian multi-camera surveillance
- Kettnaker, Zabih
- 1999
(Show Context)
Citation Context ...spatially-smooth similarity function, Bhattacharyya coefficient, face tracking. 1 INTRODUCTION REAL-TIME object tracking is the critical task in many computer vision applications such as surveillance =-=[44]-=-, [16], [32], perceptual user interfaces [10], augmented reality [26], smart rooms [39], [75], [47], object-based video compression [11], and driver assistance [34], [4]. Two major components can be d... |

92 | Real-time closed-world tracking
- Intille, Davis, et al.
- 1997
(Show Context)
Citation Context ...DUCTION REAL-TIME object tracking is the critical task in many computer vision applications such as surveillance [44], [16], [32], perceptual user interfaces [10], augmented reality [26], smart rooms =-=[39]-=-, [75], [47], object-based video compression [11], and driver assistance [34], [4]. Two major components can be distinguished in a typical visual tracker. Target Representation and Localization is mos... |

92 | Active blobs
- Sclaroff, Isidoro
- 1998
(Show Context)
Citation Context ...n criterion [6]. Since the correlation is sensitive to illumination, Hager and Belhumeur [33] explicitly modeled the geometry and illumination changes. The method was improved by Sclaroff and Isidoro =-=[67]-=- using robust M-estimators. Learning of appearance models by employing a mixture of stable image structure, motion information, and an outlier process, was discussed in [41]. In a different approach, ... |

89 |
Tracking colour objects using adaptive mixture models
- McKenna, Gong
- 1999
(Show Context)
Citation Context ... 3D, nonrigid motion, was extensively analyzed in [36], [1], [30], [73]. Explicit tracking approaches of people [69] are timeconsuming and often the simpler blob model [75] or adaptive mixture models =-=[53]-=- are also employed. The main contribution of the paper is to introduce a new framework for efficient tracking of nonrigid objects. We show that by spatially masking the target with an isotropic kernel... |

83 | Spatial color indexing and applications - Huang, Kumar, et al. - 1999 |

83 | proposal distributions: Object tracking using unscented particle filter
- Rui, Chen, et al.
(Show Context)
Citation Context ...es. Chen et al. [15] used the Hidden Markov Model formulation for tracking combined with JPDAF data association. Rui and Chen proposed to track the face contour based on the unscented particle filter =-=[66]-=-. Cham and Rehg [13] applied a variant of MHF for figure tracking. The emphasis in this paper is on the other component of tracking: target representation and localization. While the filtering and dat... |

80 | Unifying Maximum Likelihood Approaches in Medical Image Registration
- ROCHE, MALANDAIN, et al.
(Show Context)
Citation Context ...on. While the filtering and data association have their roots in control theory, algorithms for target representation and localization are specific to images and related to registration methods [72], =-=[64]-=-, [56]. Both target localization and registration maximizes a likelihood type function. The difference is that in tracking, as opposed to registration, only small changes are assumed in the location a... |

75 | Color-based tracking of heads and other mobile objects at video frame rates
- Fieguth, Terzopoulos
- 1997
(Show Context)
Citation Context ...rations is 4:19 per frame. region inside the rectangle. The surface is asymmetric due to neighboring colors that are similar to the target. While most of the tracking approaches based on regions [7], =-=[27]-=-, [50] must perform an exhaustive search in the rectangle to find the maximum, our algorithm converged in four iterations as shown in Fig. 3. Note that the operational basin of attraction of the mode ... |

71 |
Pedestrian detection and tracking with night vision
- Xu, Liu, et al.
- 2005
(Show Context)
Citation Context ...ith a kernel tracker for monitoring shopping groups in stores. Yilmaz et al. [78] combined kernel tracking with global motion compensation for forward-looking infrared (FLIR) imagery. Xu and Fujimura =-=[77]-=- used night vision for pedestrian detection and tracking, where the detection is performed by a support vector machine and the tracking is kernel-based. Rao et al. [61] employed kernel tracking in the... |