#### DMCA

## Live dense reconstruction with a single moving camera (2010)

### Cached

### Download Links

Venue: | IEEE Conference on Computer Vision and pattern Recognition |

Citations: | 107 - 5 self |

### Citations

1020 | A volumetric method for building complex models from range images.
- Curless, Levoy
- 1996
(Show Context)
Citation Context ...point samples, or volumetric signed distance functions computed from the depth maps to fit a globally consistent function that is then polygonised. This class of algorithm includes optimal approaches =-=[2, 20]-=-, though is less amenable to larger scale reconstructions due to prohibitively large memory requirements for the volumetric representation of the global functions. It is noted however that recent adva... |

548 |
Zippered polygon meshes from range images,” in
- Turk, Levoy
- 1994
(Show Context)
Citation Context ...crease the possibility of using such methods in the future [18]. The second class of algorithms work directly on the partial mesh reconstructions. One of the earliest methods, zippered polygon meshes =-=[12]-=-, obtained a consistent mesh topology by Ω Figure 5. A local surface reconstruction using a total of four images. From one scene flow iteration and associated error measures (top left). A second itera... |

530 | A comparison and evaluation of multi-view stereo reconstruction algorithms.
- Seitz, Curless, et al.
- 2006
(Show Context)
Citation Context ...llows. The dense stereo literature in computer vision has long aimed at the reconstruction of continuous surfaces from sets of registered images, and there are now many examples of impressive results =-=[11]-=-. However, this work has previously been unavailable to real-time applications due to computational cost and other restrictions. Attempts at real-time monocular tracking and dense modelling to date ha... |

471 |
Parallel tracking and mapping for small AR workspaces”,
- Klein, Murray
- 2007
(Show Context)
Citation Context ...e 3D scene and are correctly clipped by occlusions. 1. Introduction Real-time monocular camera tracking (alternatively known as real-time SFM or monocular SLAM) has recently seen great progress (e.g. =-=[3, 5]-=-). However, by design such systems build only sparse feature-based scene models, optimised to permit accurate and efficient live camera pose estimation. Their application to problems where more is nee... |

439 | Real-Time Simultaneous Localisation and Mapping with a Single Camera.
- Davison
- 2003
(Show Context)
Citation Context ...e 3D scene and are correctly clipped by occlusions. 1. Introduction Real-time monocular camera tracking (alternatively known as real-time SFM or monocular SLAM) has recently seen great progress (e.g. =-=[3, 5]-=-). However, by design such systems build only sparse feature-based scene models, optimised to permit accurate and efficient live camera pose estimation. Their application to problems where more is nee... |

369 | Poisson surface reconstruction”,
- Kazhdan, Bolitho, et al.
- 2006
(Show Context)
Citation Context ...ed is crucial in obtaining an up to date base model. Globally optimal non-parametric surface fitting techniques have suffered from high computational cost in solving large, dense systems of equations =-=[13, 4]-=-. In more recent years large reductions in the computational cost of reconstruction have traded global optimality for hierarchical, coarse-to-fine solutions. In particular, radial basis functions with... |

198 | A duality based approach for realtime tv-l1 optical flow.
- Zach, Pock, et al.
- 2007
(Show Context)
Citation Context ...lising an L 1 data fidelity norm and a total variation regularisation of the solution: Eu = ∫ Ω ∫ λ|I0(x) − I1(x + u(x))|dx + Ω |∇u|dx . (11) Details of solutions to minimising (11) are given in [9], =-=[19]-=-. Here, λ provides a user-controllable trade-off between the data fidelity and regularisation terms. The TV regularisation results in solutions that allow flow field discontinuities that are present a... |

185 | Detailed real-time urban 3D reconstruction from video. In:
- Pollefeys, Nistér, et al.
- 2008
(Show Context)
Citation Context ...rahedralisation of an SFM point cloud and visibility reasoning, but in this case high accuracy is only possible for multi-planar scenes. The work closest in spirit to ours is that of Pollefeys et al. =-=[10]-=-, who constructed an advanced, real-time system targeted at the dense modelling of urban scenes from a multicamera rig mounted on a vehicle, aided if necessary by GPS and IMU data. Real-time SFM, plan... |

171 | Three-dimensional scene flow.
- Vedula, Baker
- 2005
(Show Context)
Citation Context ...tion u ′i j . If ∆xj is sufficiently small then the image displacement ∆ui j = u′i j − uij can be approximated by the first order linearisation of P i around xj. In this case ∆xj is called scene flow =-=[15]-=-: ∆u i j = J i xj ∆xj , (1) is the Jacobian for the projection function evaluwhere J i xj ated at point xj with projection parameters P i : J i xj ∂Pi ≡ ∣ ∂x ∣ xj ≡ ⎡ ⎢ ⎣ ∂P i u ∂x ∂P i v ∂x ∂P i u ∂y... |

142 | An implicit surface polygonizer.
- Bloomenthal
- 1994
(Show Context)
Citation Context ...nstructions are integrated into the global surface model and redundant vertices are trimmed. points R 3 ↦→ R, f(x) = 0. A reconstructed mesh is extracted by polygonising the function’s zero level set =-=[1]-=-. In our application, speed is crucial in obtaining an up to date base model. Globally optimal non-parametric surface fitting techniques have suffered from high computational cost in solving large, de... |

84 | Real-Time Visibility-Based Fusion of Depth Maps.
- Merrell, Akbarzadeh, et al.
- 2007
(Show Context)
Citation Context ...me system targeted at the dense modelling of urban scenes from a multicamera rig mounted on a vehicle, aided if necessary by GPS and IMU data. Real-time SFM, plane sweep stereo and 1depth map fusion =-=[6]-=- are combined to reconstruct large urban models. In the approach we propose, we take a number of stateof-the-art components from computer vision and graphics and combine them in a significantly novel ... |

76 | H.: A Globally Optimal Algorithm for Robust TV-L1 Range Image Integration
- ZACH, POCK, et al.
(Show Context)
Citation Context ...point samples, or volumetric signed distance functions computed from the depth maps to fit a globally consistent function that is then polygonised. This class of algorithm includes optimal approaches =-=[2, 20]-=-, though is less amenable to larger scale reconstructions due to prohibitively large memory requirements for the volumetric representation of the global functions. It is noted however that recent adva... |

62 | Variational Implicit Surfaces.
- Turk, O’Brien
- 2000
(Show Context)
Citation Context ...ed is crucial in obtaining an up to date base model. Globally optimal non-parametric surface fitting techniques have suffered from high computational cost in solving large, dense systems of equations =-=[13, 4]-=-. In more recent years large reductions in the computational cost of reconstruction have traded global optimality for hierarchical, coarse-to-fine solutions. In particular, radial basis functions with... |

61 | A multi-scale approach to 3d scattered data interpolation with compactly supported basis functions,”
- Ohtake, Belyaev, et al.
- 2003
(Show Context)
Citation Context ...have enabled the dense system of constraints to be made sparse. We use a state of the art multi-scale compactly supported radial basis function (MSCSRBF) technique for 3D scattered data interpolation =-=[7]-=-. This method combines some of the best aspects of global and local function approximation and is particularly suited to the sparse, approximately oriented, point cloud obtained from the SFM processin... |

54 | An improved algorithm for TV-L1 optical flow”,
- Wedel, Pock, et al.
- 2009
(Show Context)
Citation Context ...ained by the time taken to compute the flow fields. Recently, it has been demonstrated that the solution to the variational problem can be reformulated to obtain a highly optimised GPU implementation =-=[17]-=- which we utilise here. The data term in (11) is linearised and the solution is computed in a coarse to fine manner embedded in an iterative warping framework. [17] also includes photometric normalisa... |

41 | Fast and high quality fusion of depth maps.
- Zach
- 2008
(Show Context)
Citation Context ...presentation of the global functions. It is noted however that recent advances in solving variational problems efficiently on GPU hardware increase the possibility of using such methods in the future =-=[18]-=-. The second class of algorithms work directly on the partial mesh reconstructions. One of the earliest methods, zippered polygon meshes [12], obtained a consistent mesh topology by Ω Figure 5. A loca... |

35 | ProFORMA: Probabilistic Feature-based On-line Rapid Model Acquisition.
- PAN, REITMAYR, et al.
- 2009
(Show Context)
Citation Context ... been unavailable to real-time applications due to computational cost and other restrictions. Attempts at real-time monocular tracking and dense modelling to date have had limited results. Pan et al. =-=[8]-=- recently presented a system for live modelling of simple objects by tetrahedralisation of an SFM point cloud and visibility reasoning, but in this case high accuracy is only possible for multi-planar... |

19 | Reconstructing relief surfaces.
- Vogiatzis, Torr, et al.
- 2004
(Show Context)
Citation Context ...tion is precise dense correspondence between images offering sufficient baseline for triangulation, and the main novelty of our approach is the manner in which this is enabled. In a similar manner to =-=[16]-=- we utilise a coarse base surface model as the initial starting point for dense reconstruction. Our base surface permits view predictions to be made between the cameras in a local bundle, and these pr... |

13 |
Continuous Globally Optimal Image Segmentation with Local Constraints
- Unger, Pock, et al.
- 2008
(Show Context)
Citation Context ...ce image ensures that strong image boundaries, often indicators of object boundaries, are not smoothed over in the depth map, thereby providing a form of segmentation prior on the construction of D ′ =-=[14]-=-. D ′ is then triangulated and used in place of the original base model, improving view prediction and and the quality of optical flow computation, and ultimately increasing photo-consistency in the r... |

11 |
Fast total variation for computer vision
- Pock
- 2008
(Show Context)
Citation Context ...y utilising an L 1 data fidelity norm and a total variation regularisation of the solution: Eu = ∫ Ω ∫ λ|I0(x) − I1(x + u(x))|dx + Ω |∇u|dx . (11) Details of solutions to minimising (11) are given in =-=[9]-=-, [19]. Here, λ provides a user-controllable trade-off between the data fidelity and regularisation terms. The TV regularisation results in solutions that allow flow field discontinuities that are pre... |