### Citations

3625 | Compressed sensing,” - Donoho - 2006 |

2728 | Atomic decomposition by basis pursuit,
- Chen, Donoho, et al.
- 1998
(Show Context)
Citation Context ...i-th entry in the vector). This is a convex optimization problem that can be solved by linear programming methods (e.g., generic pathfollowing primal-dual algorithm) [33], also known as basis pursuit =-=[34]-=-. 3.3 Coping with Background Clutter and Occlusion Although sparse solution for the coefficient ω can be obtained by solving an l1 optimization in (5), in the context of human pose estimation we may n... |

2632 | Robust uncertainty principles: exact signal reconstruction from highly incomplete frequency information,” - Candes, Romberg, et al. - 2004 |

1513 | Near optimal signal recovery from random projections: Universal encoding strategies?” - Candes, Tao - 2006 |

966 | Sparse Bayesian learning and the relevance vector machine,”The
- Tipping
- 2001
(Show Context)
Citation Context ...rectly. An early approach of this type represented 3D pose space as a manifold that could be approximated by hidden Markov models [24]. Agarwal and Triggs advocated the relevance vector machine (RVM) =-=[25]-=- to learn this mapping where silhouette boundary points were used as features [26]. Rosales and Sclaroff used specialized maps in addition to an inverse rendering process to learn this mapping [27]. A... |

936 | Robust face recognition via sparse representation. - Wright, Yang, et al. - 2009 |

743 | The visual analysis of human movement: a survey.
- Gavrila
- 1999
(Show Context)
Citation Context ...the use of temporal information. We will focus on previous work with a similar goal and leave interested readers to consult one of the surveys for a more complete listing of work in this general area =-=[14, 15]-=-. Previous approaches to human pose estimation from a single image can be broadly categorized as model-based or model-free based. In model-based approaches a parametric model that captures the kinemat... |

720 | Gaussian Processes for Machine Learning.
- Rasmussen, Williams
- 2006
(Show Context)
Citation Context ...he reconstruction error. 4 Experimental Results We test the proposed algorithm on synthetic [26] and real [4] data sets for empirical validation. In all experiments, we use Gaussian process regressor =-=[35]-=- to learn the mapping between image features and the corresponding 3D pose parameters. We first demonstrate the proposed method is able to estimate human pose from images with occlusions. Even without... |

515 | A survey of computer vision-based human motion capture
- Moeslund, Granum
- 2001
(Show Context)
Citation Context ...stantial amount of work on the general problem of human motion capture and understanding. As such, we find it useful to place the focus of our work within the taxonomy proposed by Moedlund and Granum =-=[14]-=- whereby the field of work is presented in the categories of person detection, tracking, pose estimation and recognition. Out approach fits best into the category of pose estimation where the goal is ... |

419 |
The Representation and Matching of Pictorial Structures.
- Fischler, Elschlager
- 1973
(Show Context)
Citation Context ...hich maximal agreement between the model and the image measurements is sought. One method for this is to simultaneously detect body parts and assemble them in a bottom-up manner. Pictorial structures =-=[16]-=- presented a convenient discrete graphical form for this that can be adapted for people using an efficient dynamic programming minimization proposed by Felzenszwalb and Huttenlocher [17] and later use... |

394 |
Convex optimization. Cambridge university press,
- Boyd, Vandenberghe
- 2004
(Show Context)
Citation Context ...∑ i |ωi|, where ωi stands for the i-th entry in the vector). This is a convex optimization problem that can be solved by linear programming methods (e.g., generic pathfollowing primal-dual algorithm) =-=[33]-=-, also known as basis pursuit [34]. 3.3 Coping with Background Clutter and Occlusion Although sparse solution for the coefficient ω can be obtained by solving an l1 optimization in (5), in the context... |

261 | TRIGGS B.: Recovering 3D human pose from monocular images.
- AGARWAL
- 2006
(Show Context)
Citation Context ...t could be approximated by hidden Markov models [24]. Agarwal and Triggs advocated the relevance vector machine (RVM) [25] to learn this mapping where silhouette boundary points were used as features =-=[26]-=-. Rosales and Sclaroff used specialized maps in addition to an inverse rendering process to learn this mapping [27]. Along a different line, Shakhnarovich do not learn a regression function, but inste... |

250 | Darrell,“Fast pose estimation with parameter-sensitive hashing”,
- Shakhnarovich, Viola, et al.
- 2003
(Show Context)
Citation Context ...s to learn this mapping [27]. Along a different line, Shakhnarovich do not learn a regression function, but instead directly make use of training examples in a lookup table using an efficient hashing =-=[28]-=-. The feature space used for these types of methods, with few exceptions, is global in the sense that the features carry no information about the body region they describe. This provides a clean top-d... |

215 | Recovering Human Body Configurations: Combining Segmentation and Recognition. In: Computer Vision and Pattern Recognition.
- Mori, Ren, et al.
- 2004
(Show Context)
Citation Context ...[17] and later used in various forms by a number of researchers [18–20]. Mori et al. followed a similar line of thought, but employed “superpixels” for the task of segmenting and detecting body parts =-=[21]-=-. Sigal et al. presented a bottom-up approach in a continuous parameter space using a modified particle filter for the minimization [1]. In contrast, Taylor developed a method to invert a kinematic mo... |

208 | Efficient matching of pictorial structures,”
- Felzenszwalb, Huttenlocher
- 2000
(Show Context)
Citation Context ...al structures [16] presented a convenient discrete graphical form for this that can be adapted for people using an efficient dynamic programming minimization proposed by Felzenszwalb and Huttenlocher =-=[17]-=- and later used in various forms by a number of researchers [18–20]. Mori et al. followed a similar line of thought, but employed “superpixels” for the task of segmenting and detecting body parts [21]... |

193 | Lee,“Inferring 3-D body pose from silhouettes using activity manifold learning”,
- Elgammal, S
- 2004
(Show Context)
Citation Context ...age in execution time. Several image representations have been proposed in discriminative pose estimation algorithms such as shape context of silhouettes [6], signed-distance functions on silhouettes =-=[7]-=-, binary principal component analysis of appearance [8], and mixture of probabilistic principal component analysis on multi-view silhouettes [2]. However, silhouettes are inherently ambiguous as diffe... |

186 | Estimating Human Body Configurations using Shape Context Matching,”
- Mori, Malik
- 2002
(Show Context)
Citation Context ...an accurate labeling of joint coordinates that provides reconstruction up to a scale ambiguity [22]. This method was combined with shape-context matching in a fully automatic system by Mori and Malik =-=[23]-=-. Model-free based approaches, which include regression and example based methods, take a top-down approach to this problem and attempt to recover a mapping from image feature space to pose parameter ... |

151 | Reconstruction of articulated objects from point correspondences in a single image. Computer Vision and Image Understanding
- Taylor
- 2000
(Show Context)
Citation Context ...ilter for the minimization [1]. In contrast, Taylor developed a method to invert a kinematic model given an accurate labeling of joint coordinates that provides reconstruction up to a scale ambiguity =-=[22]-=-. This method was combined with shape-context matching in a fully automatic system by Mori and Malik [23]. Model-free based approaches, which include regression and example based methods, take a top-d... |

143 | Finding and tracking people from the bottom up. - Ramanan, Forsyth - 2003 |

126 | Probabilistic methods for finding people,” - Ioffe, Forsyth - 2001 |

114 | Discriminative density propagation for 3-D human motion estimation”,
- Sminchisescu, Kanaujia, et al.
- 2005
(Show Context)
Citation Context ...hen the solution is not unique. Conventionally, the method of least squares can be used to find an approximate solution to this case by solving minimum l2-norm solution: min ||ω||2 subject to Aω = b. =-=(3)-=- For the system Aω = b, the minimum l2-norm solution can be obtained by ˆω2 = (A T A) −1 A T b. However, the minimum l2-norm (energy) solution ˆω2 is usually dense (with many nonzero entries), thereby... |

109 | Learning to parse pictures of people, - Ronfard, Schmid, et al. - 2002 |

103 | Inferring 3D Structure with a Statistical Image-Based Shape Model
- Grauman, Shakhnarovich, et al.
- 2003
(Show Context)
Citation Context ...ns and thus makes them unsuitable for inferring the hidden poses. In contrast, discriminative methods do not assume a particular human body model, and they can be further categorized as example-based =-=[2]-=- and learning-based [3–5]. Example-based approaches store a set of training samples along with their corresponding pose descriptors. For a given test image, a similarity search is performed to find si... |

60 | Attractive people: Assembling loose-limbed models using non-parametric belief propagation. In
- SIGAL, ISARD, et al.
- 2003
(Show Context)
Citation Context ...uman pose estimation algorithms can be categorized as generative (model-based) and discriminative (model-free). Generative methods employ a known model (e.g., tree structure) based on prior knowledge =-=[1]-=-. The pose estimation process includes two parts: 1) modeling: constructing the likelihood function and 2) estimation: predicting the most likely hidden poses based on image observations and the likel... |

57 | Predicting 3d people from 2d pictures
- Sigal, Black
- 2006
(Show Context)
Citation Context ...HOG feature descriptor computed from (a). (c) Recovered feature vector by our algorithm. (d) The reconstruction error. 4 Experimental Results We test the proposed algorithm on synthetic [26] and real =-=[4]-=- data sets for empirical validation. In all experiments, we use Gaussian process regressor [35] to learn the mapping between image features and the corresponding 3D pose parameters. We first demonstra... |

53 | Learning body pose via specialized maps
- Rosales, Sclaroff
- 2001
(Show Context)
Citation Context ...M) [25] to learn this mapping where silhouette boundary points were used as features [26]. Rosales and Sclaroff used specialized maps in addition to an inverse rendering process to learn this mapping =-=[27]-=-. Along a different line, Shakhnarovich do not learn a regression function, but instead directly make use of training examples in a lookup table using an efficient hashing [28]. The feature space used... |

44 | A local basis representation for estimating human pose from cluttered images
- Agarwal, Triggs
- 2006
(Show Context)
Citation Context ...nes, discriminative approaches have the advantage in execution time. Several image representations have been proposed in discriminative pose estimation algorithms such as shape context of silhouettes =-=[6]-=-, signed-distance functions on silhouettes [7], binary principal component analysis of appearance [8], and mixture of probabilistic principal component analysis on multi-view silhouettes [2]. However,... |

39 | Soatto,“Fast human pose estimation using appearance and motion via multidimensional boosting regression”,
- Bissacco, Yang, et al.
- 2007
(Show Context)
Citation Context ...d scenarios (e.g., videos with camera motion, dynamic background, sudden illumination change, etc.). To cope with this problem, appearance features like block SIFT descriptors [9], Haar-like features =-=[10]-=-, Histogram of oriented gradients (HOG) [6, 11, 12] or bag-of-visual-words representations [13] have been proposed for pose estimation. These descriptors contain richer information than silhouette-bas... |

27 | Relevant feature selection for human pose estimation and localization in cluttered images,” in
- Okada, Soatto
- 2008
(Show Context)
Citation Context ..., dynamic background, sudden illumination change, etc.). To cope with this problem, appearance features like block SIFT descriptors [9], Haar-like features [10], Histogram of oriented gradients (HOG) =-=[6, 11, 12]-=- or bag-of-visual-words representations [13] have been proposed for pose estimation. These descriptors contain richer information than silhouette-based features, but they inevitably encode irrelevant ... |

26 |
Evaluating Example-based Pose Estimation: Experiments on the HumanEva Sets”,
- Poppe
- 2007
(Show Context)
Citation Context ..., dynamic background, sudden illumination change, etc.). To cope with this problem, appearance features like block SIFT descriptors [9], Haar-like features [10], Histogram of oriented gradients (HOG) =-=[6, 11, 12]-=- or bag-of-visual-words representations [13] have been proposed for pose estimation. These descriptors contain richer information than silhouette-based features, but they inevitably encode irrelevant ... |

23 | Fast Algorithms for Large Scale Conditional 3D Prediction
- Bo, Sminchisescu, et al.
- 2008
(Show Context)
Citation Context ...es from compressive sensing [29–32] suggest that if the solution of ω is sparse enough, then the sparsest solution can be exactly recovered via the l1-norm optimization: min ||ω||1 subject to Aω = b, =-=(5)-=- where the l1-norm sums up the absolute weights of all entries in ω (i.e., ||ω||1 := ∑ i |ωi|, where ωi stands for the i-th entry in the vector). This is a convex optimization problem that can be solv... |

15 |
Learning generative models for multi-activity body pose estimation
- Jaeggli, Koller-Meier, et al.
- 2009
(Show Context)
Citation Context ...ve been proposed in discriminative pose estimation algorithms such as shape context of silhouettes [6], signed-distance functions on silhouettes [7], binary principal component analysis of appearance =-=[8]-=-, and mixture of probabilistic principal component analysis on multi-view silhouettes [2]. However, silhouettes are inherently ambiguous as different 3D poses can have very similar silhouettes. In add... |

12 |
Bm 3 e: Discriminative density propagation for visual tracking
- Sminchisescu, Kanaujia, et al.
(Show Context)
Citation Context ...icable in many real-world scenarios (e.g., videos with camera motion, dynamic background, sudden illumination change, etc.). To cope with this problem, appearance features like block SIFT descriptors =-=[9]-=-, Haar-like features [10], Histogram of oriented gradients (HOG) [6, 11, 12] or bag-of-visual-words representations [13] have been proposed for pose estimation. These descriptors contain richer inform... |