Thanks to visit codestin.com
Credit goes to www.scribd.com

Scribd
Upload
5 views2 pages

Lecture 9 Notes

The lecture notes cover techniques for obtaining depth data using optical centers, including Visual Hull Carving and Space Carving for 3D shape reconstruction. It also discusses Structure from Motion (SfM) for creating 3D models from 2D images, and highlights properties of good features and various feature descriptors like SIFT and SURF. Additionally, it introduces Simultaneous Localization and Mapping (SLAM) as a method for mapping unknown environments while tracking position.

Uploaded by

davor Milenkoski
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as ODT, PDF, TXT or read online on Scribd
5 views2 pages

Lecture 9 Notes

The lecture notes cover techniques for obtaining depth data using optical centers, including Visual Hull Carving and Space Carving for 3D shape reconstruction. It also discusses Structure from Motion (SfM) for creating 3D models from 2D images, and highlights properties of good features and various feature descriptors like SIFT and SURF. Additionally, it introduces Simultaneous Localization and Mapping (SLAM) as a method for mapping unknown environments while tracking position.

Uploaded by

davor Milenkoski
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as ODT, PDF, TXT or read online on Scribd
You are on page 1/ 2

Lecture 9 Notes:

We always need two optical centers in order to obtain depth data


(either 2 cameras or one-proj-one-cam setup)

Visual Hull Carving / Space Carving


Requires calibrated setups
Background <-> foreground segmentation
Not very robust to noise

Visual Hull Carving


A reconstruction method that creates an approximate shape of an object by combining silhouettes from
multiple camera views

Space Carving
A technique used to reconstruct a 3D shape from multiple images taken from different viewpoints

SFM – Structure from Motion


Creating 3D models from a 2D images taken
from different angles

Properties of good features:


– Fast comparisons
– Scale invariant
– View invariant
– Lighting invariant

Feature descriptors for matching:


– SIFT
– SURF
– ORB
– FREAK

SIFT (Scale-Invariant Feature Transform)


Difference of Gaussians
Neighborhood is divided in 16 sub-blocks
8-bin orientation histograms
128-dimensional
Multi-view Stereo (MVS)
Take results of SfM
– 6DoF Poses of images
– Estimated 3D points

Simultaneous Localization and Mapping (SLAM)


A process where a robot or device creates a map of an unknown environment while simultaneously
tracking its own position within that map.

Similar idea to SFM+MVS


Smaller compute budget
No global optimization
Dynamically extracts keyframes

Output:
– Poses + sparse point cloud
– Various ‘semi-dense methods’

You might also like