## 3D Articulated Models and Multi-View Tracking with Physical Forces

Citations: | 160 - 0 self |

### BibTeX

@MISC{Delamarre_3darticulated,

author = {Quentin Delamarre and Olivier Faugeras},

title = {3D Articulated Models and Multi-View Tracking with Physical Forces},

year = {}

}

### Years of Citing Articles

### OpenURL

### Abstract

this article we focus on the study of the gestures of a person, but the same methodology could be applied to the study of robots motions or of other kinds of articulated objects. Some examples of applications are listed in the table 1.

### Citations

1262 | Three-Dimensional Computer Vision — A Geometric Viewpoint - Faugeras |

578 | The visual analysis of human movement: A survey
- Gavrila
- 1999
(Show Context)
Citation Context ...tation of the body, but also the pose of the legs and the arms. For this purpose we use an articulated 3D model of human. Many recent model-based methods have been proposed as described in [1] and in =-=[6]-=-. They depend upon the amount of precision required by the applications and upon the number of cameras. Gavrila and Davis [7] use a best-first local 1 This work was partially supported by the European... |

479 | Iterative Point Matching for Registration of Free-Form Curves and Surfaces
- Zhang
- 1994
(Show Context)
Citation Context ...nside direction (see fig. 5). If the two contours don't intersect each other, a different kind of forces has to be applied. One solution to this problem is the ICP algorithm (Iterative Closest Points =-=[18]-=-). In its simplest form, points are chosen on one curve. Then, at each step, and until convergence, a force f i similar to those created by a spring is created between each of these points and its clo... |

451 | Computational Geometry - Preparata, Shamos - 1985 |

401 | Articulated Body Motion Capture by Annealed Particle Filtering - Deutscher, Blake, et al. - 2000 |

370 | Tracking people with twists and exponential maps
- Bregler, Malik
- 1998
(Show Context)
Citation Context ...t local 1 This work was partially supported by the European Project ESPRIT ltr 23.515: IMPROOFS (www.esat.kuleuven.ac.be/konijn/improofs.html) search after a segmentation of images. Bregler and Malik =-=[3]-=- use twists and exponential maps to model the orthographic projection of the 3D model, in order to perform a local search for pose estimation. Most of the methods using a 3D model for tracking have so... |

306 | Human motion analysis: A review
- Aggarwal, Cai
- 1999
(Show Context)
Citation Context ...n and orientation of the body, but also the pose of the legs and the arms. For this purpose we use an articulated 3D model of human. Many recent model-based methods have been proposed as described in =-=[1]-=- and in [6]. They depend upon the amount of precision required by the applications and upon the number of cameras. Gavrila and Davis [7] use a best-first local 1 This work was partially supported by t... |

231 | Model-based vision: a program to see a walking person - Hogg - 1983 |

205 | Model-based tracking of self-occluding articulated objects
- Rehg, Kanade
- 1995
(Show Context)
Citation Context ...f this 3D model to the contours. 2.1. Input data A single camera is too constraining: depth cannot be well estimated (except if you have a very precise model of what you are tracking, as described in =-=[13]-=-, [10] or [4] for the hand), the background generates false matches, occlusions are hard to manage, etc. However, two points of view of an object yield 3D information that is easier to recover. If the... |

173 | Towards model-based recognition of human movements in image sequence - Rohr |

135 | Model-based image analysis of human motion using constraint propagation - O’Rourke, Badlet - 1980 |

128 | A real time system for robust 3-D voxel reconstruction of human motions - Cheung, Kanade, et al. - 2000 |

114 | 3D human body model acquisition from multiple views. ICCV'95
- Kakadiaris, Metaxas
(Show Context)
Citation Context ...em, the idea in our work is to create forces between the 3D model and the image contours of the moving person. These forces are applied to each rigid part of the model (whereas Kakadiaris and Metaxas =-=[8, 9]-=- use forces for estimating also a better shape of the 3D model). Since we only need to recover the state parameters of the 3D model, we use an approximate model of a human, and show that it is suffici... |

110 | Tracking Persons in Monocular Image Sequences - Wachter, Nagel - 1999 |

87 | Singularity Analysis for Articulated Object Tracking - Morris, Rehg - 1998 |

72 | Computing differential properties of 3-D shapes from stereoscopic images without 3-D models - Devernay, Faugeras - 1994 |

61 | Combining Stereo and Monocular Information to Compute Dense Depth Maps that Preserve Depth Discontinuities - Fua - 1997 |

52 | A PDE-based level-set approach for detection and tracking of moving objects
- Paragios, Deriche
- 1998
(Show Context)
Citation Context ...tive contours. The contours evolve according to a PDE that incorporates optical flow and intensity measures and converge towards the silhouette of the moving person (see fig 3). For more details, see =-=[11]-=- and [12]. 2.4. How to make the images of the 3D model converge towards the contours? We assume that the initial pose of the person is known. The initialization has been done manually in the first fra... |

48 | Tracking through Singularities and Discontinuities by Random Sampling - Deutscher, North, et al. - 1999 |

47 |
Dynamics of multibody systems
- Roberson, Schwertassek
- 1988
(Show Context)
Citation Context ...ting all the constraints of the joints! Fortunately, the dynamics of systems of rigid objects has been addressed before by the robotics community [17]. Schwertassek and Roberson explain in their book =-=[14, 15]-=- how to compute the accelerations of articulated objects in time O(N), N being the number of rigid parts (20 in our case). They show how we can simplify the computation of the motion by taking into ac... |

38 |
Non-rigid matching using demons
- Thirion
- 1996
(Show Context)
Citation Context ...ed version of the images of the person. In each image, the contour of the projected 3D model is moved towards the interior normals to the real contour. More precisely, we use a technique developed in =-=[16]-=-: the Maxwell's demons. When points chosen on the real contour are within the projected contour, a force is created directed along the normal, to move the 3D model in the inside direction (see fig. 5)... |

35 | Unifying boundary and region-based information for geodesic activetracking
- Paragios, Deriche
- 1999
(Show Context)
Citation Context ...ours. The contours evolve according to a PDE that incorporates optical flow and intensity measures and converge towards the silhouette of the moving person (see fig 3). For more details, see [11] and =-=[12]-=-. 2.4. How to make the images of the 3D model converge towards the contours? We assume that the initial pose of the person is known. The initialization has been done manually in the first frame for th... |

25 |
3D model based tracking of humans in action: a multiview approach
- Gavrila, Davis
- 1996
(Show Context)
Citation Context ...y recent model-based methods have been proposed as described in [1] and in [6]. They depend upon the amount of precision required by the applications and upon the number of cameras. Gavrila and Davis =-=[7]-=- use a best-first local 1 This work was partially supported by the European Project ESPRIT ltr 23.515: IMPROOFS (www.esat.kuleuven.ac.be/konijn/improofs.html) search after a segmentation of images. Br... |

24 |
Vision based hand modeling and tracking for virtual teleconferencing and telecollaboration
- Kuch, Huang
- 1995
(Show Context)
Citation Context ... 3D model to the contours. 2.1. Input data A single camera is too constraining: depth cannot be well estimated (except if you have a very precise model of what you are tracking, as described in [13], =-=[10]-=- or [4] for the hand), the background generates false matches, occlusions are hard to manage, etc. However, two points of view of an object yield 3D information that is easier to recover. If the camer... |

21 | Finding pose of hand in video images: A stereo-based approach
- Delamarre, Faugeras
- 1998
(Show Context)
Citation Context ...f the object of interest can be in general estimated. We have studied previously the case of two cameras with a small baseline and vergence and computed a 3D reconstruction of the tracked object (see =-=[5]-=- about hand tracking). Here we tackle the alternative case where the vergence angles are more than 45 degrees and the baseline is about five meters: fig. 1 shows an example of simultaneous images take... |

21 | Gesture Estimation and Model Refinement using Monocular Camera – Ambiguity Limitation by Inequality Constraints - Hand - 1998 |

20 |
Model based estimation of 3D human motion with occlusion based on active multiviewpoint selection. CVPR'96
- Kakadiaris, Metaxas
(Show Context)
Citation Context ...em, the idea in our work is to create forces between the 3D model and the image contours of the moving person. These forces are applied to each rigid part of the model (whereas Kakadiaris and Metaxas =-=[8, 9]-=- use forces for estimating also a better shape of the 3D model). Since we only need to recover the state parameters of the 3D model, we use an approximate model of a human, and show that it is suffici... |

14 | Gomtrie algorithmique. Ediscience international - Boissonnat, Yvinec - 1995 |

13 |
TotalCalib: a fast and reliable system for off-line calibration of images sequences
- Bougnoux, Robert
- 1997
(Show Context)
Citation Context ...e cameras. See also the mpeg sequence "Movie 1". Figure 1. Three points of view of a running man. See Movie 1. The cameras are calibrated through the software package TOTALCALIB developed in=-= our lab ([2]-=-) The three sequences of images have been synchronized manually. 2.2. 3D model of a person The 3D model consists of truncated cones (arms and legs), spheres (neck, joints and head) and right parallele... |

13 | Computing di erential properties of 3-D shapes from stereoscopic images without 3-D models - Devernay, Faugeras - 1994 |

12 | Systeme de miroirs pour la stereoscopie - Mathieu, Devernay - 1995 |

11 | Articulated-Pose Estimation using Brightness- and DepthConstancy Constraints - Covell, Rahimi, et al. - 2000 |

5 | Singularities in articulated object tracking with 2-D and 3-D models - Rehg, Morris - 1997 |

3 | Totalcalib: a fast and reliable system for ooe-line calibration of images sequences - Bougnoux, Robert - 1997 |

3 | Geodesic active regions for tracking - Paragios, Deriche - 1999 |

2 |
Modelisation de la main pour sa localisation dans une sequence d'images
- Delamarre
- 1996
(Show Context)
Citation Context ...l to the contours. 2.1. Input data A single camera is too constraining: depth cannot be well estimated (except if you have a very precise model of what you are tracking, as described in [13], [10] or =-=[4]-=- for the hand), the background generates false matches, occlusions are hard to manage, etc. However, two points of view of an object yield 3D information that is easier to recover. If the cameras are ... |

2 |
Aspects of efficient and reliable multibody systems simulation
- Schwertassek, Rulka
- 1990
(Show Context)
Citation Context ...ting all the constraints of the joints! Fortunately, the dynamics of systems of rigid objects has been addressed before by the robotics community [17]. Schwertassek and Roberson explain in their book =-=[14, 15]-=- how to compute the accelerations of articulated objects in time O(N), N being the number of rigid parts (20 in our case). They show how we can simplify the computation of the motion by taking into ac... |

2 |
Dynamics of Systems of Rigid Objects
- Wittenburg
- 1977
(Show Context)
Citation Context ...to invert a 120x120 matrix at each iteration, not counting all the constraints of the joints! Fortunately, the dynamics of systems of rigid objects has been addressed before by the robotics community =-=[17]-=-. Schwertassek and Roberson explain in their book [14, 15] how to compute the accelerations of articulated objects in time O(N), N being the number of rigid parts (20 in our case). They show how we ca... |

2 | Computing dioeerential properties of 3D shapes from stereoscopic images without 3-D models - Devernay, Faugeras |

2 | gesture estimation and model re nement using monocular camera - ambiguity limitation by inequality constraints - Hand - 1998 |

1 | and Fr#d#ric Devernay. Syst#me de miroirs pour la st#r#oscopie - Mathieu - 1995 |

1 | Aspects of eOEcient and reliable multibody systems simulation - Schwertassek, Rulka |

1 | and Frédéric Devernay. Système de miroirs pour la stéréoscopie - Mathieu - 1995 |

1 | Totalcalib: a fast and reliable system for o-line calibration of images sequences - Bougnoux, Robert - 1997 |

1 | Aspects of ecient and reliable multibody systems simulation - Schwertassek, Rulka |

1 | gesture estimation and model re using monocular camera { ambiguity limitation by inequality constraints - Hand - 1998 |