Monocular SLAM and Real-Time Scene Perception

Published on 2012-10-0915498 Views

Andrew Davison

We have seen great advances in real-time 3D vision in recent years, enabled by algorithmic improvements, the continuing increase in commodity processing power and better camera technology. Research in

BMVC 2012 - Surrey

Related categories

Presentation

Monocular SLAM and Real-Time Scene Perception00:00

Robot Vision00:30

Simultaneous Localisation and Mapping (1)02:43

Simultaneous Localisation and Mapping (2)04:11

Simultaneous Localisation and Mapping (3)04:14

Simultaneous Localisation and Mapping (4)04:16

Simultaneous Localisation and Mapping (5)04:17

Simultaneous Localisation and Mapping (6)04:19

SLAM with First Order Uncertainy Propagation04:46

SLAM Using Vision: First Steps05:50

SLAM Using Active Stereo Vision06:39

Monocular SLAM07:05

MonoSLAM10:22

Application: HRP-2 Humanoid at JRL, AIST, Japan14:46

HRP2 Loop Closure15:10

Application: Wearable Robot SLAM16:16

General Components of a Scalable SLAM System17:04

SLAM as a Bayesian Network18:59

PTAM: Parallel Tracking and Mapping20:09

Real-Time Monocular SLAM: Why Filter?21:24

Double Window Optimisation for Constant Time Visual SLAM24:27

Towards Live Dense Reconstruction26:41

Solving Inverse Problems with Variational Optimisation29:52

GPGPU Processing Power32:45

Live Dense Reconstruction with a Single Camera34:07

Spherical Mosaicing using Whole Image Alignment38:03

DTAM: Dense Tracking and Mapping39:39

Cost volume data term40:35

Using all possible frames from the live camera40:52

Tracking using the dense model40:54

Why Not Increase Frame-Rate to Track Even Faster Motion?43:00

Scalable Active Matching45:12

Experimental Investigation with `Photo-Realistic' Video45:21

Experiment assuming perfect lighting47:56

Moderate lighting49:23

Real-Time Dense Surface Mapping and Tracking49:52

Signed Distance Function surface representations51:20

SDF Fusion (1)52:19

SDF Fusion (2)52:46

SDF Fusion (3)52:48

SDF Fusion (4)52:50

Real-time Surface Fusion using a Single RGB Camera57:44

Towards Greater Physical Understanding58:52