Sign inSign up
Factor Graphs for Fast and Scalable 3D Reconstruction and Mapping thumbnail
Pause
Mute
Subtitles
Playback speed
0.25
0.5
0.75
1
1.25
1.5
1.75
2
Full screen

Factor Graphs for Fast and Scalable 3D Reconstruction and Mapping

Published on Apr 03, 20143291 Views

Frank Dellaert

Simultaneous Localization and Mapping (SLAM) and Structure from Motion (SFM) are important and closely related problems in robotics and vision. I will show how both SLAM and SFM instances can be posed

BMVC 2013 - Bristol

Related categories

Computer VisionMachine Learning

Chapter list

Optimization in Factor Graphs for Fast and Scalable 3D Reconstruction and Mapping00:00
Georgia PhD program00:11
Monte Carlo Localization00:54
SLAM = Simultaneous Localization and Mapping01:47
2013: Full 3D LIDAR Mapping02:39
Large-Scale Structure from Motion03:30
3D Models from Community Databases - 103:50
3D Models from Community Databases - 204:10
4D Reconstruction - 104:48
4D Reconstruction - 205:43
3D Reconstruction06:12
4D Structure over Time06:39
Outline - 107:32
Outline - 207:45
Acknowledgements08:34
Boolean Satisfiability09:19
Constraint Satisfaction Problems - 110:27
Constraint Satisfaction Problems - 210:40
Graphical Models11:10
Continuous Probability Densities12:18
Simultaneous Localization and Mapping (SLAM)12:58
Factor Graph -> Smoothing and Mapping (SAM)! - 113:39
Factor Graph -> Smoothing and Mapping (SAM)! - 214:15
Structure from Motion14:40
Outline - 315:31
Gaussian Factor Graph == mxn Matrix15:52
Linear Least Squares - 116:38
Linear Least Squares - 216:45
Square Root Factorization17:27
Intel Dataset18:08
MIT Killian Court Dataset18:19
Variable Elimination - 118:44
Variable Elimination - 220:56
Small Example - 121:33
Small Example - 222:33
Small Example - 322:45
Small Example - 422:46
Small Example - 522:47
Small Example - 623:32
Elimination at work - 123:51
Elimination at work - 224:03
Outline - 425:39
GTSAM 225:52
Elimination order is again CRUCIAL26:36
Approximate Minimum Degree27:22
Divide and Conquer27:45
St. Peter’s Basilica, Rome28:28
Nested Dissection (but good separators are hard)29:11
Nested-Dissection for SFM: Hyper-SFM29:39
GTSAM in Robotics30:23
iSAM [Kaess et al., TRO 08] - 130:57
iSAM [Kaess et al., TRO 08] - 231:46
iSAM2: Bayes Tree for Manhattan Sequence33:51
Notable Applications of iSAM - 135:18
Notable Applications of iSAM - 235:23
Notable Applications of iSAM - 335:47
Mapping Aircraft Carriers etc ...36:02
Long-­‐term Visual Mapping38:02
Reduced Pose Graph – 10 Floors38:42
Reduced Pose Graph39:34
Very long-term Mapping40:09
Outline - 541:08
Of Eiffel Towers ... - 142:06
Of Eiffel Towers ... - 243:02
Iterative Methods44:44
Main Idea (with Viorela Ila and Doru Balcan)45:31
Linear Reparametrization is Preconditioning46:42
Original, max47:09
Preconditioner47:47
Preconditioned, max48:23
Illustration with Real Data48:41
Notre Dame49:19
Piazza San Marco49:25
Results49:41
Outline - 649:56
GTSAM 350:42
Backbone Methods51:58
Who needs correspondence ...52:41
Distributed SLAM (and SFM?)53:38
GTSAM: SAM + MPC54:01
Outline - 754:37