#### DMCA

## DTAM: Dense Tracking and Mapping in Real-Time

### Cached

### Download Links

Citations: | 128 - 5 self |

### Citations

2232 | Nonlinear total variation based noise removal algorithms
- Rudin, Osher, et al.
- 1992
(Show Context)
Citation Context ...together, enforcing ξ = α as θ → 0, resulting in the original energy (6). As a function of ξ, the convex sum g(u)‖∇ξ(u)‖ɛ + Q(u) is a small modification of the TV-L 2 2 ROF image denoising model term =-=[11]-=-, and can be efficiently optimised using a primal-dual approach [1][16][3]. Also, although still non-convex in the auxiliary variable α, each (7)Figure 3. Incremental cost volume construction; we sho... |

692 | Lucas-Kanade 20 years on : A unifying framework. part I : The quantity approximated, the warp update rule and the gradient descent approximation
- Baker, Matthews
- 2004
(Show Context)
Citation Context ...etween Iv and the live image Il to estimate Tlv, and hence the true pose Twl = TwvTvl. We parametrise an update to ˆ Tvl by ψ ∈ R 6 belonging to the Lie Algebra se3 and define a forward-compositional =-=[2]-=- cost function relating photometric error to changing parameters: F (ψ) = 1 2 ∑ u∈Ω ( ) 2 fu (ψ) = 1 2 ‖f(ψ)‖2 2, (19) ( ( fu(ψ) = Il π KTlv(ψ)π −1 (u, ξv (u)) )) − Iv (u) (20) ( 6∑ ) Tlv(ψ) = exp i=1... |

461 |
Parallel tracking and mapping for small AR workspaces
- Klein, Murray
- 2007
(Show Context)
Citation Context ...keyframe inverse depth map (4th from left). In comparison to the nearly 300 × 10 3 points estimated in our keyframe, we show the ≈ 1000 point features used in the same frame for localisation in PTAM (=-=[6]-=-). Estimating camera pose from such a fully dense model enables tracking robustness during rapid camera motion. Q(u) + λC (u, α(u)) is now trivially point-wise optimisable and can be solved using an e... |

425 | A first-order primal-dual algorithm for convex problems with applications to imaging
- Chambolle, Pock
- 2011
(Show Context)
Citation Context ...s a function of ξ, the convex sum g(u)‖∇ξ(u)‖ɛ + Q(u) is a small modification of the TV-L 2 2 ROF image denoising model term [11], and can be efficiently optimised using a primal-dual approach [1][16]=-=[3]-=-. Also, although still non-convex in the auxiliary variable α, each (7)Figure 3. Incremental cost volume construction; we show the current inverse depth map extracted as the current minimum cost for ... |

106 | Live dense reconstruction with a single moving camera
- Newcombe, Davison
- 2010
(Show Context)
Citation Context ...e, and ERC Starting Grant 210346. We are very grateful to our colleagues at Imperial College London for countless useful discussions. tems working with a hand-held camera have recently appeared (e.g. =-=[9, 13]-=-), but these still rely on feature tracking. Here we present a new algorithm, DTAM (Dense Tracking and Mapping), which unlike all previous real-time monocular SLAM systems both creates a dense 3D surf... |

85 | Improving the agility of keyframebased SLAM
- Klein, Murray
- 2008
(Show Context)
Citation Context ...true rotation is not strictly rotational (Figure 6). A similar step is performed before feature matching in PTAM’s tracker, computing first the interframe 2D image transform and fitting a 3D rotation =-=[7]-=-. The rotation estimate helps inform our current best estimate of the live camera pose, ˆ Twl. We project the dense model in to a virtual camera v at location Twv = ˆ Twl, with colourFigure 6. MSE co... |

82 | Robust multi-sensor image alignment
- Irani, Anandan
- 1998
(Show Context)
Citation Context ... tracking. Although section 2.3.2 describes how we can handle local illumination changes whilst tracking, we are not robust to real-world global illumination changes that can occur. Irani and Anandan =-=[5]-=- showed how a normalised cross correlation measure can be integrated into the objective function for more robustness to local and global lighting changes. As an alternative to this, in future work, we... |

74 | Fast cost-volume filtering for visual correspondence and beyond, CVPR, pp
- Rhemann, Hosni, et al.
- 2011
(Show Context)
Citation Context ... the set of frames nearby and overlapping r, to compute the values stored in the cost volume. the cost volume (called a disparity space image in stereo matching [14], and generalised more recently in =-=[10]-=- for any discrete per pixel labelling) stores the accumulated photometric error as a function of inverse depth d. The average photometric error Cr(u, d) is computed by projecting a point in the volume... |

40 |
Large displacement optical flow computation without warping
- Steinbrücker, Pock
- 2009
(Show Context)
Citation Context ...image space as in [13], all images used in the data term must be kept increasing computational cost as more overlapping images are used. Instead, following the large displacement optic flow method of =-=[12]-=- we approximate the energy functional by coupling the data and regularisation terms through an auxiliary variable α : Ω → R, ∫ Eξ,α = Ω { g(u)‖∇ξ(u)‖ɛ + 1 (ξ(u) − α(u))2 2θ } + λC (u, α(u)) du . The c... |

40 | Fast and high quality fusion of depth maps, in
- Zach
- 2008
(Show Context)
Citation Context ...le to get more complete, accurate and robust results using dense methods which make use of all of the data in an image. Methods for high quality dense stereo reconstruction from multiple images (e.g. =-=[15, 4]-=-) are becoming real-time capable due to their high parallelisability, allowing them to track the currently dominant GPGPU hardware curve. Meanwhile, the first live dense reconstruction sysThis work wa... |

39 | Some first-order algorithms for total variation based image restoration
- Aujol
- 2009
(Show Context)
Citation Context ... (6). As a function of ξ, the convex sum g(u)‖∇ξ(u)‖ɛ + Q(u) is a small modification of the TV-L 2 2 ROF image denoising model term [11], and can be efficiently optimised using a primal-dual approach =-=[1]-=-[16][3]. Also, although still non-convex in the auxiliary variable α, each (7)Figure 3. Incremental cost volume construction; we show the current inverse depth map extracted as the current minimum co... |

28 | Sampling the Disparity Space Image
- Szeliski, Scharstein
- 2004
(Show Context)
Citation Context ...ames indexed as m ∈ I(r), where I(r) is the set of frames nearby and overlapping r, to compute the values stored in the cost volume. the cost volume (called a disparity space image in stereo matching =-=[14]-=-, and generalised more recently in [10] for any discrete per pixel labelling) stores the accumulated photometric error as a function of inverse depth d. The average photometric error Cr(u, d) is compu... |

27 |
Real-time dense geometry from a handheld camera
- Stuehmer, Gumhold, et al.
(Show Context)
Citation Context ...e, and ERC Starting Grant 210346. We are very grateful to our colleagues at Imperial College London for countless useful discussions. tems working with a hand-held camera have recently appeared (e.g. =-=[9, 13]-=-), but these still rely on feature tracking. Here we present a new algorithm, DTAM (Dense Tracking and Mapping), which unlike all previous real-time monocular SLAM systems both creates a dense 3D surf... |

15 | 3d reconstruction using an n-layer heightmap
- Gallup, Pollefeys, et al.
- 2010
(Show Context)
Citation Context ...le to get more complete, accurate and robust results using dense methods which make use of all of the data in an image. Methods for high quality dense stereo reconstruction from multiple images (e.g. =-=[15, 4]-=-) are becoming real-time capable due to their high parallelisability, allowing them to track the currently dominant GPGPU hardware curve. Meanwhile, the first live dense reconstruction sysThis work wa... |

14 | Real-time spherical mosaicing using whole image alignment
- Lovegrove, Davison
- 2010
(Show Context)
Citation Context ...e true solution. We use a coarse-fine strategy over a power of two image pyramid for efficiency and to increase our range of convergence. 2.3.1 Pose Estimation We first follow the alignment method of =-=[8]-=- between consecutive frames to obtain rotational odometry at lower levels within the pyramid, offering resilience to motion blur since consecutive images are similarly blurred. This optimisation is mo... |

5 |
Fast Numerical Algorithms for Total Variation Based Image Restoration
- Zhu
- 2008
(Show Context)
Citation Context ...). As a function of ξ, the convex sum g(u)‖∇ξ(u)‖ɛ + Q(u) is a small modification of the TV-L 2 2 ROF image denoising model term [11], and can be efficiently optimised using a primal-dual approach [1]=-=[16]-=-[3]. Also, although still non-convex in the auxiliary variable α, each (7)Figure 3. Incremental cost volume construction; we show the current inverse depth map extracted as the current minimum cost f... |