#### DMCA

## I.: Dense reconstruction using 3d object shape priors

Venue: | In: CVPR, IEEE |

Citations: | 13 - 1 self |

### Citations

1396 | Object detection with discriminatively trained part-based models.
- Felzenszwalb, Girshick, et al.
- 2010
(Show Context)
Citation Context ...ly acquires new depth meshes at key-frames and fuses these into a global volumetric representation, in parallel we run a part-based object-class detector based on the effective procedure described in =-=[9]-=-. Although this cannot proceed at frame-rate, it does not need to. Rather, we process the most recent key-frame with the detector. If an object is found, further processing ensues (see below) but if n... |

1011 | A Volumetric Method for Building Complex Models from Range Images,
- Curless, Levoy
- 1996
(Show Context)
Citation Context ... to the surface is a weighted average of all previous measurements, helping to smooth out errors. Since in almost all respects our implementation closely follows the prior art, we refer the reader to =-=[5]-=- for further details. 3. Incorporating object knowledge Our objective in this paper is to show how the ability to detect objects and incorporate them into a SLAM map is beneficial, as a step towards a... |

654 | Eigentracking: Robust matching and tracking of articulated objects using a view-based representation,”
- Black, Jepson
- 1998
(Show Context)
Citation Context ...t surface is equal to the probability that the back projected 3D point lies on the zero-level of the SDF. We compute this probability as the negative exponential of the robust German-McClure function =-=[3]-=-: P (X|Φ,Ωv) = e− Φ(X0) 2 σ+Φ(X0)2 (9) This probability equals one when a depth pixel is backprojected onto the zero-level of the SDF and decreases monotonically at a rate controlled by σ (we used σ =... |

466 |
Parallel tracking and mapping for small AR workspaces.
- Klein, Murray
- 2007
(Show Context)
Citation Context ... camera trajectory is unknown, the joint on-line estimation problem for scene structure and camera pose has become known as visual Simultaneous Localisation and Mapping. Early methods for visual SLAM =-=[7, 11]-=- concentrated on accurate camera pose estimation using only sparse reconstructions. These have various disadvantages, such as their inability to provide occlusion information. More recently, with the ... |

271 | KinectFusion: Real-time dense surface mapping and tracking.
- Newcombe, Davison, et al.
- 2011
(Show Context)
Citation Context ...frommultiple mesh depth estimates at a single point is lost. To address this limitation we fuse the local depth maps into a dense volumetric parametrization of the 3D world 12879 akin to that used in =-=[26, 10, 16]-=-. The volume is discretised as a 3D grid of voxels, each containing two components: (i) F , the value of a signed distance function representing an approximation of the distance from the voxel to the ... |

223 | Probabilistic non-linear principal component analysis with gaussian process latent variable models,
- Lawrence
- 2005
(Show Context)
Citation Context ...llow analogously to those wrt. pose and scale, by replacing ∂Φ v ∂Xvn ∂Xvn ∂λp with ∂Φ v ∂lq and ∂Φ∂λp with ∂Φ ∂lq . These final two derivatives are the ones of the standard GP-LVM generative process =-=[13]-=-, on which the inverse DCT transform has been applied. 4. Map update Once the shape and pose estimation of the object has converged (as measured using the standard LevenbergMarquardt test), we fuse th... |

166 | A shape-based approach to the segmentation of medical imagery using level sets
- Tsai, Yezzi, et al.
(Show Context)
Citation Context ... the most effective and popular approaches to represent shape knowledge is to use dimensionality reduction to capture the shape variance as low dimensional latent shape spaces. Initial works, such as =-=[24]-=-, focused on (implicitly or explicitly defined) 2D shapes, and used linear dimensionality reduction in the form of principal component analysis (PCA). More recent works use nonlinear dimensionality re... |

129 | Dtam: Dense tracking and mapping in real-time.
- Newcombe, Lovegrove, et al.
- 2011
(Show Context)
Citation Context ...vantages, such as their inability to provide occlusion information. More recently, with the introduction of new GPU-based computation devices, real-time, dense SLAM has become a technical possibility =-=[10, 15]-=-. These dense approaches use the raw image information (such as colour or gradients) to estimate the scene geometry, leveraging weak priors such as scene smoothness in the absence of image texture, an... |

76 | H.: A Globally Optimal Algorithm for Robust TV-L1 Range Image Integration
- ZACH, POCK, et al.
(Show Context)
Citation Context ...frommultiple mesh depth estimates at a single point is lost. To address this limitation we fuse the local depth maps into a dense volumetric parametrization of the 3D world 12879 akin to that used in =-=[26, 10, 16]-=-. The volume is discretised as a 3D grid of voxels, each containing two components: (i) F , the value of a signed distance function representing an approximation of the distance from the voxel to the ... |

74 | Fast cost-volume filtering for visual correspondence and beyond, CVPR, pp
- Rhemann, Hosni, et al.
- 2011
(Show Context)
Citation Context ... mirroring the approach of [15], we formulate the initial depth map estimation problem as one of finding the depth of each point that is seen in one reference image. Following various prior art (e.g. =-=[21]-=-) [15] assumes known camera intrinsic and extrinsics and uses photo-consistency as a proxy for depth; viz, a point on a lambertian surface in the scene should project to the same colour in all the fra... |

53 |
Robust real-time visual tracking using pixel-wise posteriors. In:
- Bibby, Reid
- 2008
(Show Context)
Citation Context ...defined foreground and background regions, as a function of the projected 3D SDF Φ, using the functions Pf and Pb from eq (4) in each reference view v. This measure, first proposed for 2D tracking by =-=[2]-=-, is in contrast to the full random forest classfier used in [19] to obtain image statistics. Here we can use these much weaker (and therefore much faster to compute) image statistics because of the a... |

43 | Semantic structure from motion
- Bao, Savarese
(Show Context)
Citation Context ...end to appear in the reconstruction. Humans deal with these issues via the use of high level semantic information. In an incremental move towards such high-level representations, works such as [4] or =-=[1]-=- use known objects as features in the SLAM system. While this has the effect of increasing SLAM robustness and accuracy, both approaches are limited by their use of a sparse map and by the fact that t... |

27 | Detection-based object labeling in 3d scenes
- Lai, Bo, et al.
- 2012
(Show Context)
Citation Context ...gulate a rough object centre in 3D by back-projecting the centre of the detection in two different views. We thus require a detection in at least two key-frames before proceeding to estimate the pose =-=[12]-=-. This of course incurs a delay, but does also help eliminate some false positives, since these are less likely to persist across frames than true positives. Next, we estimate the vertical axis of the... |

22 |
Towards Simultaneous Recognition, Localization and Mapping for Hand-held and Wearable Cameras
- Castle, Gawley, et al.
- 2007
(Show Context)
Citation Context ...facts tend to appear in the reconstruction. Humans deal with these issues via the use of high level semantic information. In an incremental move towards such high-level representations, works such as =-=[4]-=- or [1] use known objects as features in the SLAM system. While this has the effect of increasing SLAM robustness and accuracy, both approaches are limited by their use of a sparse map and by the fact... |

16 | Online 3d reconstruction using convex optimization
- Graber, Pock, et al.
- 2011
(Show Context)
Citation Context ...vantages, such as their inability to provide occlusion information. More recently, with the introduction of new GPU-based computation devices, real-time, dense SLAM has become a technical possibility =-=[10, 15]-=-. These dense approaches use the raw image information (such as colour or gradients) to estimate the scene geometry, leveraging weak priors such as scene smoothness in the absence of image texture, an... |

15 | Robust 3d pose estimation and efficient 2d region-based segmentation from a 3d shape prior
- Dambreville, Sandhu, et al.
- 2008
(Show Context)
Citation Context ...plicitly defined) 2D shapes, and used linear dimensionality reduction in the form of principal component analysis (PCA). More recent works use nonlinear dimensionality reduction such as Kernel PCA in =-=[6]-=- and Gaussian Process Latent Variable Models (GP-LVM) in [17]. This led to 3D shape priors being first introduced in [23]. Most recently, [19] learn GP-LVM latent spaces of 3D shapes and use them in m... |

13 |
A Bayesian Estimation of Building Shape Using MCMC
- Dick, Torr, et al.
(Show Context)
Citation Context ...both approaches are limited by their use of a sparse map and by the fact that they consider the objects to be of fixed and perfectly known shape. A more generic semantic reconstruction is proposed in =-=[8]-=-, where shape and layout priors of buildings are learned offline. Once several plans of a building are found in several images, the model that best matches the priors and images is estimated. While th... |

9 |
RGB-D camera-based parallel tracking and meshing
- Lieberknecht, Huber, et al.
- 2011
(Show Context)
Citation Context ...roach is used. 2.2. Robust Map Representation The process above can be repeated for several reference frames, and the resulting depth maps merged into a single global map. While [15] and others (e.g. =-=[14]-=-) simply accumulate the corresponding meshes, this has several disadvantages: first, it does not give any occupancy information even though this information is fundamental for many applications; secon... |

7 |
Shared shape spaces
- Prisacariu, Reid
- 2011
(Show Context)
Citation Context ...pressed using the Discrete Cosine Transform, retaining the n lowest frequencies in the DCT. The resulting space of DCT coefficients is then used to train a GP-LVM with a low-dimensional latent space. =-=[18, 19]-=- showed that this representation could more effectively capture useful within-class shape variation, and exclude out-of-class variation, than other competing methods 1Though the trimaps could be learn... |

6 |
A unified energy minimization framework for model fitting in depth
- Ren, Reid
- 2012
(Show Context)
Citation Context ... This probability equals one when a depth pixel is backprojected onto the zero-level of the SDF and decreases monotonically at a rate controlled by σ (we used σ = 100). This approach was also used in =-=[20]-=-, in which depth data coming from a Microsoft Kinect unit was used for simultaneous model based 3D tracking and calibration. Unlike [20] however, here we (i) also make use of the RGB image data, (ii) ... |

2 |
A nonrigid kernel-based framework for 2D-3D pose estimation and 2D image segmentation
- Sandhu, Dambreville, et al.
(Show Context)
Citation Context ... More recent works use nonlinear dimensionality reduction such as Kernel PCA in [6] and Gaussian Process Latent Variable Models (GP-LVM) in [17]. This led to 3D shape priors being first introduced in =-=[23]-=-. Most recently, [19] learn GP-LVM latent spaces of 3D shapes and use them in monocular simultaneous 2D segmentation, 3D reconstruction and 3D pose recovery. Our objective in this paper is to address ... |

1 |
Nonlinear Shape Manifolds as Shape
- Prisacariu, Reid
(Show Context)
Citation Context ...reduction in the form of principal component analysis (PCA). More recent works use nonlinear dimensionality reduction such as Kernel PCA in [6] and Gaussian Process Latent Variable Models (GP-LVM) in =-=[17]-=-. This led to 3D shape priors being first introduced in [23]. Most recently, [19] learn GP-LVM latent spaces of 3D shapes and use them in monocular simultaneous 2D segmentation, 3D reconstruction and ... |

1 |
LibPaBOD: A library for part-based object detection
- Rodriguez-Molina, Marin-Jimenez
(Show Context)
Citation Context ...y-frame with the detector. If an object is found, further processing ensues (see below) but if not, the process retrieves the most recent key-frame and tries again. Using our implementation (based on =-=[22]-=-) we can process a single 320 × 240 frame in 2.1s on average. The use of this object class detector has several advantages in our context: it can detect object-classes that exhibit considerable in-cla... |